Why a British 'Silicon Valley outsider' is one of AI's biggest players

0:02Sky News, the full story first.

0:14He's won a Nobel Prize, transformed medical research and is trying to create artificial general intelligence. Demis Hassabis is the Brit that you've never heard of who could already have changed your life. This is why.

0:35Hello, Ed Conway here. Here is what might sound like a slightly random question. What do AirPods, bananas and a pair of shoddily made running trainers from the 1970s have in common? On the surface, not very much. But on my podcast, Stuff Matters, I will uncover the deeper stories behind each of them. You will see how together they tell a fascinating new story about the world we live in. The first full series of Stuff Matters is out now. Just search for it on your podcast app and follow.

1:12Hi everyone, Neil here. And over the summer, we are going to be delving into some of the less well-known titans of the tech world. And we start with an absolute doozy. Chess prodigy, video game designer, neuroscientist, entrepreneur, Nobel Prize winner. it's the sort of CV that makes you actively question your own life. So Demas Isabis is the British co-founder of Google DeepMind and one of the most important figures in artificial intelligence. His company's AlphaFold system has transformed our understanding of proteins, opening up exciting new possibilities in the world of medicine, but also earning him a share of the Nobel Prize for Chemistry.

1:51My passion is always to build these types of tools to help us with scientific discovery. And I think we'll get amazing things out of that, cures for diseases, helping with energy, climate. A lot of the big challenges we have in front of us as humanity today. But that is only part of his ambition. Hassabis wants to create artificial general intelligence, a machine capable, in essence, of learning across practically any task. So who is the man pursuing that goal? And why does he see opportunity in AI where others see only danger? I sat down with Sebastian Malaby, who spent three years with unprecedented access to Hassabis for his book, The Infinity Machine.

2:32Sebastian, I mean, Demis Hassabis is a remarkable man by any measure, but the signs were there from an early age. I mean, this was a child prodigy, was it not? It was, yeah. He climbed up on a chair to watch his dad playing chess when he was about four years old. and just by looking at the game, he could understand it. And then within a few weeks, he was defeating adults and within a year or so, he was playing tournaments. And then by the time he was 10 or 11, he was captain of the chess team in Britain, the junior chess team. Best in Britain, but second best in the world. That came out of nowhere.

3:07You are speaking to someone who played board two for the Scottish primary school's chess team, thankfully a lot longer in the past than I would have had to come up against Demis. So I understand the utility of chess as a game for regulating your thinking, about teaching you how to apply yourself in situations demanding critical thinking. I mean, this wasn't just a one-off. I mean, academically, he was incredibly strong as well. Yeah, I mean, he dropped out of school for a whole year at one point to just focus on chess and taught himself what he needed to know in his spare time in his bedroom. He also was a coder.

3:41He used his winnings from the chess tournaments to buy himself home computers and began to write code, write a program to play games like you try chess. And I think his system choked because it didn't have enough memory. But he then played another game, which he coded. He got into Cambridge University to read computer science when he was just 16 out of a state school in North London. So yeah, definitely very able in many different departments. Yeah, it's the polymath quality, or at least a variety of skills, which could perhaps put under one umbrella, but which took him in different directions.

4:14Chess on the one hand, he also mentioned coding there. I mean, it wasn't just a bit of a bedroom activity. I mean, he was writing games and incredibly successful ones at a very young age. He got into Cambridge really young and then the university said, look, you're academically ready, but you probably would have a better time if you waited till you were nearer to 18 to show up on campus because you have more friends and stuff. So go do something else in the meantime. So he got himself a job at a video game design company called Bullfrog, which was a bit of an iconic studio for video games back in the day.

4:46And he co-wrote, co-designed a game called Theme Park, which then sold millions of copies. So when he showed up at Cambridge, half of his cohort had actually played the game that he had written. and the revenues from that game were so extreme that his boss, Peter Molyneux, tried to persuade him to not do Cambridge by writing him a cheque. I worked out the value of the cheque in kind of today's money. It was well over a million dollars. And Demis just said, I'm not cashing that cheque. I prefer to go and study computer science. And what do we know then about his time at university? I mean, was he as devoted to his studies or was he all these other extracurricular activities that he started at such a young age?

5:26Well, he continued to play chess for the Queen's College team. And that's where he took up the game of Go, the ancient Chinese game. And he also got very serious about table football, believe it or not. And I thought that was a bit of a joke and didn't take it seriously. And he said, no, no, no, I was, Sebastian, I was the best player in the whole of Cambridge. I'm like, whatever, you know, yeah, yeah, yeah. No, no, no, you don't understand. You don't understand. I've watched the videos, this is Dem is talking. I've watched the videos of the professional players on YouTube and they don't know the snake shot where you kind of curl your arm around the lever and you get extra friction and you get extra power on the shot.

6:02Okay, okay. So I beat everybody that came from. He's so competitive. So he would do all these things. He was also fascinated not only by his own subject, computer science, but also by biology. He had a mate in the bar who said, there's this Nobel Prize conjecture in 1976 by the Nobel Prize winner Christian Amvinsen, who said you could devise or divine the structure, the shape of proteins in nature if you just knew the code on the string that makes them up, the kind of DNA code. And he always thought about that, put it in the back of his mind, and then, of course, later on, that was the basis for him winning a Nobel Prize.

6:41We can assume of the man at that point. I mean, as you say, there is a competitiveness to the man, to the then young man. But there was also an aspect of creativity as well. I just think taking all of those together, you could kind of almost see the direction that he was about to go into in terms of his career. I think he's a classic case of somebody who has what they call range. He's not somebody who specializes in something and puts in the famous 10 ,000 hours of practice to become world class. He was much more eclectic. He called himself a magpie. He would collect ideas and knowledge from all over the place.

7:15Creativity was part of it. He was pretty good at music. His dad was a wannabe professional musician, never quite made it. I think one of his siblings is a musician. So he had that whole side to him. And also visual design, I think, is something he cared a lot about. And in the design of the video game he did well at, that was also part of what he brought to the table. And he had this extraordinary ambition to found his own company. I mean, at Cambridge in 1997, when he graduated, nobody was going to found their own company. They were going to do a PhD, a graduate degree, they might join a big company, whatever.

7:50I mean, I talked to his contemporaries and they just said to me, look, it wasn't a thing. Nobody else had that idea, but Demis always was designing his own path forward. It sounds like when he left Cambridge, he had a very clear idea of what he was at least going to attempt to try and do. And in terms of his professional life, so much when we think of Demis Hesibis is Google DeepMind. So just explain what exactly it is. So what was the most extraordinary thing about him of all was that he had this ambition from the time he designed that video game that he wanted to build powerful artificial intelligence.

8:23And then he read a book called Gödel-Escher Bach, which is quite hard to describe. It's a super long, dense, complicated book. But basically, the idea that he took from it was that the human brain works on ones and zeros, kind of electrical signals in the brain. And that's sort of the same as a computer. If you could make a computer as big as the human brain, basically the human brain and the computer would be very similar. They could do similar things. They could be similarly intelligent. So he regarded human intelligence as the existence proof that meant that AI would one day be possible. And this is crazy because back in the mid-1990s, you know, AI couldn't do anything.

9:00It couldn't recognize the photograph of a cat. And yet he was convinced it would one day work. And so when he left Cambridge, he founded a video game company called Elixir with the idea that he would start to work on AI, smuggle the AI into the video game. And the revenues from the video game would be his way of starting his path, which is kind of his life's mission at that point. Just in terms, though, of his pursuit of AI, I can see a fascination with systems. I can see through his interest in biology, a fascination with understanding the kind of the neurological processes. was to push forward with artificial intelligence, you know, simply about creating an AI model.

9:41Or was it kind of running counter to most of the AI we see these days? Was it about understanding the human mind through replicating it in AI, rather than, as some of the companies out there today seem to be trying to do, replacing the human mind with the artificial mind? He was definitely motivated by scientific curiosity. But also, his idea was that there were mysteries in theoretical physics, and all around the world, there were scientific mysteries that he wanted to unravel. And his idea was that the only way to do better than Isaac Newton or his other scientific heroes was to have a better machine to help you.

10:17And artificial intelligence would be the ultimate scientific machine. So when DeepMind was founded, what was its purpose? What was the ultimate goal? Well, it was founded in 2010, and the goal was solve intelligence in order to solve everything else. Pretty ambitious. I mean, when he says everything else, he means physics, biology, chemistry, climate science, you know, end of climate change. Basically, he means everything. Despite the many, many aspects of science that he is pursuing, but it was the Nobel Prize for Chemistry. Explain exactly how AI, how Hussibis himself kind of took AI, utilised it and won the ultimate prize for many in science.

10:57The notion was that nature is built out of proteins, which are like little mini factories in your body or in plants or whatever. And these proteins have very, very intricate shapes, beautiful, delicate kind of filigreed shapes. And the shape starts actually, though, from just one string that you could stretch out into a straight line, which is an amino acid. And on that strand of amino acid, there is DNA code. And the notion from the 1970s was that if you analyze the code in the DNA, you could predict the way that that strand of amino acid would fold itself up like a self-executing origami model into this intricate shape.

11:43And in 2016, Demis Hassabis decided that now was the moment where DeepMind was sophisticated enough at artificial intelligence that he could enter that field of predicting protein shapes and actually solve it. not just be good at it, but be so accurate in predicting the shapes of the proteins that you would no longer need to discover the shapes through a complicated process called X-ray crystallography, where it would take five years of a scientist's career to figure out one of these protein shapes. And so Dem is set to work. He got a team at DeepMind to work on this. And by 2018, they'd made some progress.

12:23They were now the best in the world, but they were not accurate enough in the prediction to obsolete x-ray crystallography and actually the boss of that team at DeepMind was called Andrew Senior and he had the view, he said to Demis, you know, we're the best in the world but we're not going to solve this, it's too difficult, let's just declare victory and move on and Demis said no, no, no, no, no, no, no, this is not the point, the point is we need to really advance science by being able to predict the shape so accurately that you could build, let's say, a new medicine. So they pushed ahead and by the end of 2020, they solved it.

13:03And then the really sort of gangster move on the end of all this is that Demis decides, okay, we're going to put this online, all these results for free. So any scientist anywhere in the world who needs to know the precise shape of a protein, something which used to take five years of scientific work to figure out, now you just like do a Google search and you can find that precise shape online for free. And so everywhere I go, whenever I meet a medical researcher, I ask them about this and they say, oh yeah, yeah, we use that several times a week. And it's really changed the nature of structural biology.

13:39Clearly, it was not going to be enough. A Nobel Prize would not have been enough for Demis to head to the hills and enter early retirement, clearly. His work with artificial intelligences is not yet over. And Alpha Genome, another project, is one that fascinated me. For those who haven't heard about it, just explain. So Alpha Genome takes the idea further and it's kind of building on the protein folding discovery to apply it to other areas of biology and genomics and just kind of create a suite of products which are going to make medical discovery faster. We should say that, you know, for now, there's no medicine that has been discovered because of AI which is in people's bodies because the clinical trial process takes quite a long time.

14:25The breakthrough on AlphaFold was the end of 2020. It's been six years or five and a half years. We're not quite there yet in terms of normal people feeling the result, but I think that it will come. Am I wrong in the description of the way in which Demis Sibis is looking to use, to utilize artificial intelligence, that he sees it as a collaborator, as someone with whom you work, rather than a replacement for those people doing the jobs right now? I think that question is always a complicated one. Of course, certain tasks that humans do are replaced by machines. And that's a story that's as old as technology, right?

15:05I mean, frankly, I have difficulty finding my way around a city unless I get my phone out and look at the map, right? I'm still thinking hard about lots of different things. My brain is not going to sleep, but it's just outsourced one task. And I think, you know, artificial intelligence is going to do that also. And it's up to individuals to use it in a way that isn't causing you to go to sleep, but it's just actually extending what you can do. And I think that's how Demis would see it. And that's how all of our listeners should see it too, that it's up to you. It's up to you to use it in a way that is helpful and boosts you and makes you think harder, not the opposite.

15:42See, that all sounds very positive. But I imagine that there are some people out there in the world who take a look, who squint at what Demisus Ibus is actually doing and come to the conclusion that what he's actually pursuing is the creation of the world's smartest scientist. And in creating that, that does rather imply that humanity might not be worth quite as much after that process. Well, look, I'll give you an analogy. So, you know, any scientist in the 20th century was standing on the shoulders of previous scientists. and leveraging what those other scientists had discovered before them to try to push forward on the boundary.

16:20And I think the optimistic story would be that same with AI. Of course we want other people's ideas. Any individual scientist, by the way, who goes to the lab anywhere in the world is benefiting from colleagues and co-authors and collaborators who have ideas that that individual didn't have, but you kind of bounce ideas off other people and that's how you make progress in a lot of fields. And so why wouldn't we bounce ideas off the AI as well? I don't think it has to displace you. Now, maybe a different question would be not for the frontier scientists, but for sort of the average clerical worker in a company.

16:59And there I think replacement is more of a threat. And that could be brutal. And I think that, you know, the government should be doing more about policies that would help people to find a different job, would maybe cushion the shock of losing a job. So I take all that very seriously. But actually, I think at the frontier that you were asking about of scientific discovery, I think more smart brains, including artificial brains, are probably a good thing.

17:33Hello, Ed Conway here. Here's what might sound like a slightly random question. What do AirPods, bananas and a pair of shoddily made running trainers from the 1970s have in common? On the surface, not very much But on my podcast Stuff Matters I will uncover the deeper stories behind each of them You will see how together they tell a fascinating new story about the world we live in The first full series of Stuff Matters is out now Just search for it on your podcast app and follow

18:09For about three years, you spoke to him pretty regularly. What is he like as a man? So the striking thing about Demis is that he's very approachable, relatable, easy to understand. And right from the first time I met him, I went to a tech conference and he was there. And, you know, you see this boyish figure dressed very simply with kind of sneakers and a round neck sweater and kind of perky and smiley and comes up and chats and is very kind of unpretentious. Then he gets on the stage and he starts riffing about neuroscience, computer science, biology, chemistry, physics, history of music, philosophy, you name it.

18:46I mean, the ideas are just explosively varied and eclectic. But the manner is super relatable and basically nice. He married his girlfriend from Cambridge. They have two children. It's all very grounded. And that's what's great about him. You know, when I talked to him for two hours at a time in a pub in North London, in an upstairs room that nobody knew about, he would talk about so many different things in such an understandable, relatable way. I'm not a scientist, but I never felt kind of left behind. And, you know, that's amazing. I would say, though, there's one sort of qualification which I think is worth putting in the mix, just because it's sort of actually helpful to all of us as people, which is that even the most impressive people I ever meet, And I think Demis is right up there as probably number one.

19:32They do have blind spots, right? All human beings have blind spots. None of us are perfect. And so there were a few things that Demis had a blind spot about. I said, you know, you're quite controlling. And I mean that in the sense that he's so clever that whenever he debates anybody about anything, in the nicest possible way, he's just thought about it more. He's cleverer. He's got more arguments. He's more lucid. Yeah, he wins the argument and he dominates the discussion. It's not like he can't help it, but he is controlling. And you say, oh, I'm not controlling because my mother taught me it's very bad to control people and she's very religious and she had these values and I take that very seriously.

20:11And I think all of that is sincerely meant like he doesn't enjoy the idea of controlling people, but he does control them. And so I think that's a lesson about, you know, the limitations of individuals. And we all could do with both a bit of charity in the way that we are judged. None of us is perfect. And maybe also a bit of checks and balances and everybody benefits from having people around them who will point out when they have a blind spot. One of the things that really stands out about Demis is the fact that, you know, he has always been kind of a Silicon Valley outsider. I mean, geographically as much as anything else.

20:48He was born and bred in London, the child of immigrant parents, a mother who came from Singapore, Singapore Chinese, and a father who is by origin Greek, Cypriot. But that kind of makes him the perfect figure of the London melting pot, a super international city. He feels very kind of loyal to Britain. And when he raised money, he was under pressure to move to California because the funders for DeepMind, you know, the first one was Peter Thiel, a famous California venture capitalist. And Peter Thiel wanted him to move to California. He said, no, I'm staying in London. And then, you know, Google bought him.

21:26Google is headquartered in Silicon Valley. Demis had no interest in moving to California. and it was almost, I think it was a condition of the sale that he would be able to stay in London, run DeepMind and his team would stay in London. He does that because he's patriotic, right? I mean, he actually likes the values of Britain and he thinks that America is too unequal, too harsh, too sort of, you know, hype driven, fake it till you make it, move fast and break things. He doesn't like all that stuff. He's a deep scientist and he wants to do things in a way that benefits humanity and he's quite serious about that.

22:02Serious to the point that he reflects very deeply on the potential negative impacts of artificial intelligence in general. I mean, you write in your briefs Haunted by Oppenheimer, the kind of the father of the atomic bomb. What is it about the story of the atomic bomb that resonates with him, do you think? You know, it's partly that Robert Oppenheimer in his glory moments was leading this team of amazing scientists out into the desert in New Mexico. And everybody focused like a laser on the scientific challenge of building the atom bomb and the outcome of the Second World War was going to maybe depend on it.

22:43And they just were completely heads down on a fascinating scientific challenge and they cracked it. So that's a sort of scientific hero story which appeals deeply, deeply to Demis. That energy and single-minded devotion to achieving a mission is really deeply something that he resonates with. And so I think that's one way in which he identifies with Oppenheimer. But there is also, as you say, the warning side, which is that Oppenheimer regretted the atom bomb after it was used the second time in Nagasaki. And he tried to persuade the US government to hand the technology over to the United Nations and to sort of demilitarize it in a way.

23:22And Demis has been, from the get-go, very concerned about the safety, the downside risks of artificial intelligence when he first met his future co-founder, Shane Legg. It was at a safety lecture in London where Shane Legg said that by 2030, computers would be so powerful that artificial intelligence would compete with humans and maybe threaten humans. And that's what was the founding kernel of DeepMind. So weirdly, you know, by agreeing on a safety risk, they also agreed that they were going to build this technology that would cause the safety risk. There's something about scientists who just, they can't resist the sweetness of discovery, as they say, and yet they're worried by it.

24:05And Oppenheimer very much embodies that duality of both determination to push forward with science and also fear of the consequences of that determination. So what does the future look like for Demis Isibis? I mean, what is he ultimately now trying to achieve beyond, I mean, beyond the achievements that he already has in his CV? Well, as I said, he's very competitive and he would like a second Nobel Prize. Who wouldn't? There's a big, very interesting question for Demis in the next kind of five, maybe five, 10 years, which is, you know, one path for him, you know, he's running Google DeepMind.