Podcast Episode Summary
The podcast episode "Possible" features an interview with Demis Hassabis, the co-founder and CEO of DeepMind, a leading AI research organization. The main topic of the episode is the future of AI and its potential to transform various industries, including medicine, energy, and sustainability.
Key Points:
Takeaways:
Participants:
Guest Demis Hassabis:
Hosts:
Reid Hoffman
Not all journalism is the same. Take The Guardian. Our coverage has something unique. Fierce independence. Nobody owns us or tells us what we can and can't say. So we're free to report the whole picture. We connect what's happening in Washington to the rest of the globe. Expose corruption wherever we find it. And give fresh perspective on everything. From wellness and soccer to culture, the climate and more. Read, watch and listen to The Guardian for free at theguardian.com.
These days, every business leader is under pressure to save money, but you can't crush the competition just by cutting costs. To win, you need to spend smarter and move faster. You need Brex. Brex is the intelligent finance platform that breaks the trade-off between control and speed with smart corporate cards, high-yield banking, and AI-powered expense management. Join the 30,000 companies that spend smarter and move faster with Brex. Learn more at brex.com slash grow.
Game theory, medicine, multi-modality, the nature of innovation, and how board games and video games shape our understanding of the future of AI. Enjoy the episode and remember, you can find it and subscribe to Possible wherever you listen to podcasts.
AI is going to affect the whole world. It's going to affect every industry. It's going to affect every country. It's going to be the most transformative technology ever, in my opinion. So if that's true, and it's going to be like electricity or fire, then I think it's important that the whole world participates in its design. I think it's important that it's not just 100 square miles of patch of California. I do actually think it's important that we get these other inputs, the broader inputs, not just geographically, but also different subjects, philosophy, social sciences, economists, not just the tech companies, not just the scientists involved in deciding how this gets built and what it gets used for.
Aria Finger
Hi, I'm Reid Hoffman.
Reid Hoffman
And I'm Aria Finger.
Aria Finger
We want to know how, together, we can use technology like AI to help us shape the best possible future. With support from Stripe, we ask technologists, ambitious builders, and deep thinkers to help us sketch out the brightest version of the future. And we learn what it'll take to get there. This is possible.
Reid Hoffman
In the 13th century, Sir Galahad embarked on a treacherous journey in pursuit of the elusive Holy Grail. The Grail, known in Christian lore as the Cup Christ used in the Last Supper, had disappeared from King Arthur's table. The Knights of the Round Table swore to find it. After many trials, Galahad's pure heart allowed him the unique ability to look into the Grail and observed divine mysteries that could not be described by the human tongue.
In 2020, a team of researchers at DeepMind successfully created a model called AlphaFold that could predict how proteins will fold. This model helped answer one of the holy grail questions of biology. How does a long line of amino acids configure itself into a 3D structure that becomes the building block of life itself? In October 2024, three scientists involved with AlphaFold won a Nobel Prize for these efforts.
This is just one of the striking achievements spearheaded by our guest today. Demis Hassabis is a British artificial intelligence researcher, co-founder, and CEO of the AI company DeepMind. Under his leadership, DeepMind developed AlphaGo, the first AI to defeat a human world champion in Go, and later created AlphaFold, which solved the 50-year protein folding problem. He is considered one of the most influential figures in AI. Reid and I sat down for an interview with Demis, in which we talked about everything from game theory to medicine to multimodality and the nature of innovation and creativity. Here's our conversation with Demis Hassabis.
Demis, welcome to Possible. It was awesome dining with you at Queen's. It was kind of a special moment in all kinds of ways. I think I'm going to start with a question that came from your Babbage Theater lecture and also from the Fireside Chat that you did with Mohamed El-Erian. Share with us the moment where you went from thinking chess is the thing that I have spent my childhood doing to what I want to do is start thinking about thinking. I want to accelerate the process of thinking and that computers are a way to do that. How did you arrive at that? What age were you? What was that turn into metacognition?
Demis Hassabis
Well, first of all, thanks for having me on the podcast. Chess for me is where it all started, actually in gaming. I started playing chess when I was four, very seriously, all through my childhood, playing for most of the England junior teams, captaining a lot of the teams. For a long while, my main aim was to become a professional chess player, a grandmaster, maybe one day, possibly a world champion. That was my whole childhood, really. Every spare moment, not at school, I was going around the world playing chess against adults in international tournaments.
Around 11 years old, I sort of had an epiphany. Although I love chess, and I still love chess today, is it really something that one should spend their entire life on? Is it the best use of my mind? That was one thing that was troubling me a little bit. But the other thing was, as we were going to training camps with the England chess team, we started to use early chess computers to try and improve your chess. I remember thinking that we were supposed to be focusing on improving the chess openings and chess theory and tactics. But actually, I was more fascinated by the fact that someone had programmed this inanimate lump of plastic to play very good chess against me. I was fascinated by how that was done. I really wanted to understand that and then eventually try and make my own chess programs.
Aria Finger
It's so funny. I was saying to Reid before this, my seven-year-old school just won the New York State chess championship. So they have a long way to go before they get to you. But he takes it on faith like, "Oh, yeah, mom, I'm just going to go play chess kid on the computer. I'll go play against the computer a few games," which was a revelation decades ago. I remember when I was in middle school, it was obviously the Deep Blue versus Gary Kasparov. This was a man versus machine moment. One thing that you've gestured at about this moment is that it illustrated, based on grandmaster data, it was like brute force versus a self-learning system. Can you say more about that dichotomy?
Demis Hassabis
Well, first of all, it's great your son's playing chess. I think it's fantastic. I'm a big advocate for teaching chess in schools as a part of the curriculum. I think it's fantastic training for the mind, just like doing maths or programming would be. It certainly affected the way I approach problems and problem solve and visualize solutions and plan. It teaches you all these amazing meta skills dealing with pressure. You learn all of that as a young kid, which is fantastic for anything else you're going to do.
As far as Deep Blue goes, you're right. Most of these early chess programs, and then Deep Blue became the pinnacle of that, were these types of expert systems. At the time, that was the favored way of approaching AI, where the programmers solve the problem, in this case playing chess, and then they encapsulate that solution in a set of heuristics and rules, which guides a kind of brute force search towards making a good chess move. I was fascinated by these early chess programs that they could do that, but I was also slightly disappointed by them.
By the time it got to Deep Blue, I was already studying at Cambridge in my undergrad. I was actually more impressed with Kasparov's mind because I'd already started studying neuroscience than I was with the machine. The machine was this brute of a machine. All it can do is play chess. Kasparov can play chess at the same sort of roughly the same level, but also can do all the other amazing things that humans can do. I thought, doesn't that speak to the wonderfulness of the human mind? More importantly, it means something was very fundamentally missing from Deep Blue and these expert system approaches to AI.
Even though it was a pinnacle of AI at the time, Deep Blue did not seem intelligent. What was missing was its ability to learn new things. For example, it was crazy that Deep Blue could play chess to world champion level, but it couldn't even play tic-tac-toe. You'd have to reprogram. Nothing in the system would allow it to play tic-tac-toe. That's odd. That's very different to a human grandmaster, who could obviously play a simpler game trivially. It was not general in the way that the human mind is. I think those are the hallmarks of intelligence, and they were needed if we wanted to crack AI.
Aria Finger
Go a little bit into the deep learning, which is part of the reason why DeepMind was named where it is. Part of that seemingly contrarian hypothesis that you guys played out with self-play and a learning system was that this learning approach was the right way to generate these significant systems. Say a little bit about having the hypothesis, what the trek through the desert looked like, and then what finding the Nile ended up with.
Demis Hassabis
We started DeepMind in 2010 before anyone was working on this in industry, and there was barely any work on it in academia. We partially named the company DeepMind, the deep part, because of deep learning. It was also a nod to Deep Thought in Hitchhiker's Guide to the Galaxy and Deep Blue and other AI things. But it was mostly around the idea we bet on these learning techniques. Deep learning and hierarchical neural networks had just been invented in seminal work by Jeff Hinton and colleagues in 2006. It was very, very new. Reinforcement learning, which has always been a speciality of DeepMind, is the idea of learning from trial and error, learning from your experience, and then making plans and acting in the world. We combined those two things. We sort of pioneered doing that and we called it deep reinforcement learning. Deep learning to build a model of the environment or what you were doing, in this case a game, and then reinforcement learning to do the planning and the acting and actually accomplish it, to build agent systems that could accomplish goals. In the case of games, that is maximizing the score, winning the game.
We felt that that was actually the entirety of what's needed for intelligence. The reason we were pretty confident about that is from using the brain as an example. Basically, those are the two major components of how the brain works. The brain is a neural network. It's a pattern matching and structure finding system, but then it also has reinforcement learning and this idea of planning and learning from trial and error and trying to maximize reward. In the human brain and the animal brain, the mammal brain, the dopamine system implements that, a form of reinforcement learning called TD learning.
That gave us confidence that if we pushed hard enough in this direction, even though no one was really doing that, that eventually this should work, because we have the existence proof of the human mind. That's why I also studied neuroscience. When you're in the desert, you need any source of water or any evidence that you might get out of the desert. Even a mirage in the distance is a useful thing to understand in terms of giving you some direction when you're in the midst of that desert. AI was itself in the midst of that because several times this had failed. The expert system approach had reached a ceiling.
I could easily hog the entire interview, so I'm trying not to.
Aria Finger
One of the things that the learning system created was solving what was previously considered an insoluble problem. There were even people who thought that classical computational techniques couldn't solve Go. But it did. Not only did it solve Go, but in the classic Move 37, it demonstrated originality, creativity that was beyond the thousands of years of Go play and books and the hundreds of years of very serious play. What was that moment of Move 37 like for understanding where AI is? And what do you think the next Move 37 is?
Demis Hassabis
The reason Go was considered to be and ended up being so much harder than chess, so it took another 20 years, even us with AlphaGo, and all the approaches that have been taken with chess, these expert systems approaches had failed with Go. They couldn't even beat a professional, let alone a world champion. The reason was two main reasons. One is the complexity of Go is so enormous. One way to measure that is there are 10 to the power 170 possible positions, far more than atoms in the universe. There's no way you can brute force a solution to Go. It's impossible. But even harder than that is that it's such a beautiful, esoteric, elegant game.
It's considered an art form in Asia. It's both beautiful aesthetically, but it's all about patterns rather than brute calculation, which chess is more about. Even the best players in the world can't really describe to you very clearly what are the heuristics they are using. They just kind of intuitively feel the right moves. They'll sometimes just say, "Why did you play this move? Well, it felt right." It turns out their intuition, if they're a brilliant player, their intuition is brilliant, fantastic, and it's an amazingly beautiful and effective move. But that's very difficult to encapsulate in a set of heuristics and rules to direct how a machine should play Go.
That's why all of these Deep Blue methods didn't work.
We got around that by having the system learn for itself what are good patterns, what are good moves, what are good motifs and approaches, and what are valuable and high probability of winning positions. It learned that for itself through experience, through seeing millions of games and playing millions of games against itself. That's how we got AlphaGo to be better than world champion level. The additional exciting thing about that is that it means those kinds of systems can actually go beyond what we as the programmers or the system designers know how to do. No expert system can do that because it's strictly limited by what we already know and can describe to the machine.
But these systems can learn for themselves. That's what resulted in Move 37 in game two of the famous world championship match we had against Lee Sedol in Seoul in 2016. That was a truly creative move. Go has been played for thousands of years. It's the oldest game humans have invented and it's the most complex game. It's been played professionally for hundreds of years in places like Japan. Even despite all of that exploration by brilliant human players, this Move 37 was something never seen before. Worse than that, it was thought to be a terrible strategy. If you go and watch the documentary, which I recommend on YouTube, of AlphaGo, you'll see the professional commentators nearly fell off their chairs when they saw Move 37 because they thought it was a mistake.
They thought that the computer operator, Aja, had misclicked on the computer because it was so unthinkable that someone would play that. In the end, it turned out 100 moves later, that Move 37, the piece that was put down on the board, was in exactly the right place to be decisive for the whole game. Now it's studied as a great classic of the history of Go, that game and that move.
That's exactly what we hoped these systems would do. The whole point of my motivation, my whole life of working on AI, was to use AI to accelerate scientific discovery. Those kinds of new innovations, albeit in a game, are what we were looking for from our systems. I think that is an awesome rendition of why it is these learning systems are, even now, doing original discovery.
Aria Finger
What do you think the next Move 37 might be for opening our minds to the way that AI can add a whole lot to the quality of human thought, human existence, human science?
Demis Hassabis
I think there will be a lot of Move 37s in almost every area of human endeavor. The thing I've been focusing on since then is mostly how can we apply those types of AI techniques, those learning techniques, those general learning techniques to science at big areas of science. I call them root node problems. Problems where if you think of the tree of all knowledge that's out there in the universe, can you unlock some root nodes that unlock entire branches or new avenues of discovery that people can build on afterwards? For us, protein folding and AlphaFold was one of those. It was always top of my list. I have a kind of mental list of all these types of problems that I've come across throughout my life, just being generally interested in all areas of science and thinking through which ones would be both hugely impactful and suitable for these types of techniques. I think we're going to see a new golden era of these types of new strategies, new ideas in very important areas of human endeavor.
I would say one thing, though, is that we haven't fully cracked creativity yet. I don't want to claim that. I definitely describe three levels of creativity. I think AI is capable of the first two. The first one would be interpolation. You give it a million pictures of cats, and you say, "Create me a prototypical cat." It will just average all the million cat pictures that it's seen. That prototypical one won't be in the training set, so it will be a unique cat, but it's not very interesting from a creative point of view. It's just an averaging.
The second thing would be what I call extrapolation. That's more like AlphaGo. You've played 10 million games of Go, you've looked at a few million human games of Go, but then you come up with, you extrapolate from what's known to a new strategy never seen before, like Move 37. That is very valuable already. I think that is true creativity.
But then there's a third level, which I call invention or out of the box thinking. Not only can you come up with a Move 37, but could you have invented Go? Another measure I like to use is if we went back to the time of Einstein in the early 1900s, could an AI system actually come up with general relativity with the same information that Einstein had at the time? Clearly today, the answer is no to those things. It can't invent something as a game as great as Go, and it wouldn't be able to invent general relativity just from the information that Einstein had at the time. There's still something missing from our systems to get true out of the box thinking. I think it will come, but we just don't have it yet.
Aria Finger
I think so many people outside of the AI realm would be surprised that it all starts with gaming, but that's sort of gospel for what we're doing. That's how we created these systems. Switching gears from board games to video games, can you give us the elevator pitch explanation for what exactly makes an AI that can play StarCraft II, like AlphaStar, so much more advanced and fascinating than the one that can play chess or Go?
Demis Hassabis
With AlphaGo, we cracked the pinnacle of board games. Go was always considered the Mount Everest of games AI for board games. But there are even more complex games by some measures if you take on board the most complex strategy games that you can play on computers. StarCraft 2 is acknowledged to be the classic of the genre of real-time strategy games. It's a very complex game. You've got to build up your base and your units and other things. Every game is different. The board game is very fluid and you've got to move many units around in real time.
The way we cracked that was to add an additional level of a league of agents competing against each other, all seeded with slightly different initial strategies. You get a sort of survival of the fittest. You have a tournament between them all. It's a kind of multi-agent setup now. The strategies that win out in that tournament go to the next epoch. Then you generate some other new strategies around that. You keep doing that for many generations. You're having this idea of self-play that we had in AlphaGo, but you're adding in this multi-agent competitive, almost evolutionary dynamic. Eventually you get an agent or a series or set of agents that are the Nash distribution of agents.
No other strategy dominates them, but they dominate the most number of other strategies. Then you have this kind of Nash equilibrium. You pick out the top agents from that. That succeeded very well with this type of very open-ended gameplay. It's quite different from what you get with chess or Go where the rules are very prescribed and the pieces that you get are always the same. It's a very ordered game. Something like StarCraft is much more chaotic. It's interesting to have to deal with that. It has hidden information too. You can't see the whole map at once. You have to explore it. It's not a perfect information game, which is another thing we wanted our systems to be able to cope with: partial information situations.
That's actually more like the real world. Very rarely in the real world do you actually have full information about everything. Usually you only have partial information and then you have to infer everything else in order to come up with the right strategies.
Aria Finger
Part of the game side of this, I presume you've heard that there's this theory of Homo Ludens, that we're game players. Is that informing the thinking about how games is both strategic, but also a framing for science acceleration, a framing for the serendipity of innovation? In addition to the fitness function, the evolution of self-play, the ability to play, scale, compute, are there other deeper elements to the game playing nature that allows this thinking of thinking?
Demis Hassabis
I'm glad you brought up Homo Ludens. It's a wonderful book. It basically argues that games playing is actually a fundamental part of being human. In many ways, that's the act of play. What could be more human than that? It leads into creativity, fun, all of these things get built on top of that. I've always loved them as a way to practice and train your own mind in situations that you might only ever get a handful of times in real life, but they're usually very critical. What company to start, what deal to make, things like that. I think games is a way to practice those scenarios. If you take games seriously, you can actually simulate a lot of the pressures one would have in decision-making situations. Going back to earlier, that's why I think chess is such a great training ground for kids to learn, because it does teach them about all of these situations.
It's the same for AI systems too. Games were the perfect proving ground for our early AI system ideas, partly because they were invented to be challenging and fun for humans to play. There are different levels of gameplay. We could start with very simple games like Atari games, and then go all the way up to the most complex computer games like StarCraft, and continue to challenge our system. We were in the sweet spot of the S-curve. It's not too easy, it's trivial, or too hard, you can't even see if you're making any progress. You want to be in that maximum part of the S-curve where you're making almost exponential progress. We could keep picking harder and harder games as our systems got improved.
The other nice feature about games is because they're some kind of microcosm of the real world, they've usually been boiled down to very clear objective functions. Winning the game or maximizing the score is usually the objective of a game. That's very easy to specify to a reinforcement learning system or an agent-based system. It's perfect for hill climbing against and measuring ELO scores, ratings, and exactly where you are. Finally, you can calibrate yourselves against the best human players. You can sort of calibrate what your agents are doing in their own tournaments. Even with the StarCraft agent, we had to eventually challenge a professional grandmaster at StarCraft to make sure that our systems hadn't overfitted somehow to their own tournament strategies. It needed to be grounded with, "Oh, it can actually beat a genuine human grandmaster StarCraft player."
The final thing is you can generate as much synthetic data as you want with games too. That is coming into vogue right now, about data limitations and large language models and how many tokens left in the world and has it read everything in the world. For things like games, you can actually just play the system against itself and generate lots more data from the right distribution.
Aria Finger
Can you double click on that for a moment? You said it is in vogue to talk about, are we running out of data? Do we need synthetic data? Where do you stand on that issue?
Demis Hassabis
I've always been a huge proponent of simulations and AI. It's also interesting to think about what the real world is in terms of a computational system. I've always been involved with trying to build very realistic simulations of things. Now, of course, that interacts with AI because you can have an AI that learns a simulator of some real world system just by observing that system or all the data from that system.
The current debate is to do with these large foundation models now pretty much use the whole internet. Once you've tried to learn from those, what's left? That's all the language that's out there. There are other modalities like video and audio. I don't think we've exhausted all of that kind of multimodal tokens, but even that will reach some limit. The question then comes of, can you generate synthetic data? That's why you're seeing quite a lot of progress with maths and coding, because in those domains, it's quite easy to generate synthetic data. The problem with synthetic data is, are you creating data that is from the right distribution, the actual distribution? Does it mimic the kind of real distribution? And also, are you generating data that's correct?
And of course, for things like maths, for coding, and for things like gaming, you can actually test the final data and verify if it's correct right before you feed it in as input into the training data for a new system. So it's very amenable in certain areas. In fact, it turns out the more abstract areas of human thinking that you can verify and prove that it's correct. And so that unlocks the ability to create a lot of synthetic data.
Reid Hoffman
So one of the things that is also, in addition to the frequent discussion around data—how do we get more—one of the questions is, in order to do AI, is it important to actually have it embedded in the world?
Demis Hassabis
Yeah. Well, interestingly, if we talked about this five years ago, or certainly 10 years ago, I would have said that some real-world experience—maybe through robotics, when we talk about embodied intelligence, we usually mean robotics—or a very accurate simulator, like some kind of ultra-realistic game environment, would be needed to fully understand the physics of the world around you and the physical context around you. There's actually a whole branch of neuroscience predicated on this called "action in perception." This is the idea that one can't actually fully perceive the world unless you can also act in it.
The argument goes: how can you really understand the concept of the weight of something unless you can pick things up and compare them with each other? Can you really get that notion just by looking at things? It seems hard. Certainly for humans, I think you need to act in the world. Acting in the world is part of your learning; you're an active learner. In fact, reinforcement learning is like that because the decisions you make give you new experiences, but those experiences depend on the actions you took. Those are the experiences you'll then subsequently learn from. So reinforcement learning systems are involved in their own learning process because they're active learners.
I think you can make a good argument that this is also required in the physical world.
Now, it turns out I'm not sure I believe that anymore. With our systems, especially our video models—if you've seen VO2, our latest video models, completely state-of-the-art, which we released late last year—it kind of shocked even me. Even though we were building this thing, it can, by watching a lot of YouTube videos, figure out the physics of the world. There's a sort of funny Turing test for video models: can you chop a tomato? Can you show a video of a knife chopping a tomato with the fingers and everything in the right place, and the tomato doesn't magically spring back together, or the knife goes through the tomato without cutting it? VO can do it. If you think through the complexity of the physics—what you've got to keep consistent—it's pretty amazing. It's hard to argue that it doesn't understand something about physics and the physics of the world. And it's done it without acting in the world, certainly not as a robot. So it's not clear to me there is a limit now with just passive perception.
The interesting thing is that I think this has huge consequences for robotics as an embodied intelligence application. The types of models we've built—Gemini and now VO, and we'll be combining those things together at some point in the future—we've always built Gemini to be multimodal from the beginning. We still lead on all the multimodal benchmarks for two reasons. One, we have a vision for a universal digital assistant, an assistant that goes with you on digital devices but also in the real world, maybe on your phone or a glasses device, and actually helps you in the real world—recommend things, help you navigate, help with physical things like cooking.
For that to work, you obviously need to understand the context you're in. It's not just the language I'm typing into a chatbot; you have to understand the 3D world I'm living in. To be a really good assistant, you need to do that. The second thing is exactly what you need for robotics as well. We released our first big Gemini robotics work, which has caused a bit of a stir. That's the beginning of showcasing what we can do with these multimodal models that understand the physics of the world, with a little bit of robotics fine-tuning on top for the actions, motor actions, and planning a robot needs to do.
It looks like it's going to work. So now I think these general models are actually going to transfer to the embodied robotic setting without too much extra special casing, extra data, or extra effort. That's probably not what most people, even the top roboticists, would have predicted five years ago.
Reid Hoffman
That's wild. Thinking about benchmarks and what we're going to need these digital assistants to do, when we look under the hood of these big AI models, some people would say "attention." The trade-off is thinking time versus output quality. We need them to be fast, but we also need them to be accurate. Talk about that trade-off and how it's going in the world right now.
Demis Hassabis
Well, we pioneered that area of thinking systems because that's what our original gaming systems all did—Go, AlphaGo, but most famously AlphaZero, our follow-up system that could play any two-player game. There you always have to think about your time budget, your compute budget for the planning part. You can pre-train the model, just like we do with our foundation models today. You can play millions of games offline and have your model of chess or Go. But at test time, at runtime, you've only got one minute to think about your move. That's still a limited compute budget. What's very interesting today is this trade-off between using a more expensive, larger base model—like our different size models, Gemini Flash, Pro, or even bigger, Ultra—which are more costly to run and take longer, but are more accurate and capable.
You can run a bigger model with a shorter number of planning steps, or you can run a very efficient, smaller model that's slightly less powerful but you can run it for many more steps. Currently, we're finding it's roughly about equal. But of course, we want to find the Pareto frontier of that—the exact right trade-off of the size of the model and the expense of running it versus the amount of thinking time and thinking steps you can do per unit of compute time. That's fairly cutting-edge research right now that I think all the leading labs are probably experimenting on. There's not a clear answer to that yet.
Reid Hoffman
All the major labs—DeepMind, others—are all working intensely on coding assistance. There are a number of reasons. It accelerates productivity across the whole front. It has a good fitness function. It's also one of the ways everyone is going to be enhanced in productivity, having a software co-pilot agent for help. As you're building these, there's a tendency to start with computer languages designed for humans. What would be computer languages designed for an AI or an agentic world, or designed for this hybrid process of a human plus an AI? Is that a good world to start looking at? How would it change our theory of computation, linguistics, et cetera?
Demis Hassabis
I think we are entering a new era in coding. As you say, all the leading labs are pushing on this frontier for many reasons. It's easy to create synthetic data, so that's another reason everyone is pushing on this vector. I think we're going to move into a world where sometimes it's called "vibe coding," where you're basically coding with natural language. We've seen this before with computers. When I first started programming in the eighties, we were doing assembler. That seems crazy now. You start with C, then you get Python. One could see this as the natural evolution of going higher and higher up the abstraction stack of programming languages, leaving more and more of the lower-level implementation details to the compiler.
Now, one could view this as the natural final step: we just use natural language. Everything becomes a super high-level programming language. I think that's maybe what we'll get to. The exciting thing is that it will make coding accessible to a whole new range of people—creatives, designers, game designers, app writers—who normally would not have been able to implement their ideas without the help of teams of programmers. That's pretty exciting from a creativity point of view.
It may also be very good, certainly in the next few years, for coders as well. I think with these AI tools, the people who are going to get the most benefit initially will be the experts in that area who also know how to use these tools in precisely the right way, whether that's prompting or interfacing with your existing code base. There's going to be an interim period where the current experts who embrace these new tools—whether filmmakers, game designers, or coders—are going to be superhuman in terms of what they're able to do. I see that with some film director and film designer friends of mine who are able to create pitch decks for new film ideas in a day on their own.
It's a very high-quality pitch deck that they can pitch for a $10 million budget. Normally, they would have had to spend tens of thousands of dollars just to get to that pitch deck, which is a huge risk. So there's going to be a whole new incredible set of opportunities. Then there's the question of the creative arts: will there be new ways of working much more fluidly? Instead of doing Adobe Photoshop, you're co-creating this thing with a fluid, responsive tool. It could feel more like *Minority Report*, with an interface swirling around you. It will require people to get used to a very new workflow to take maximum advantage of that, but when they do, it'll probably be incredible.
They'll be 10x more productive.
Aria Finger
I want to go back to the world of multimodal that we were talking about before, with robots in the real world. Right now, most AI doesn't need to be multimodal in real time because the internet is not multimodal. For our listeners, that means absorbing many types of input—voice, text, vision—at once. Can you go deeper into what you think the benefits of truly real-time, multimodal AI will be, and what are the challenges to get to that point?
Demis Hassabis
First of all, we live in a multimodal world. We have our five sensors, and that's what makes us human. If we want our systems to be brilliant tools or fantastic assistants, in the end they're going to have to understand the world—the spatiotemporal world we live in—not just our linguistic, mathematical, abstract thinking world. They'll need to be able to act in, plan in, and process things in the real world and understand the real world. The potential for robotics is huge. I don't think it's had its ChatGPT or its AlphaFold moment yet, in science and language, or its AlphaGo moment. I think that's to come, but I think we're close. As we talked about before, the shortest path I see for that to happen is these general multimodal models eventually being good enough, maybe not very far from now, to install on a robot—perhaps a humanoid robot with cameras. There are additional challenges: you've got to fit it locally on the local chips to have fast enough latency. But as we all know, just wait a couple of years, and those state-of-the-art systems today will fit on a little mobile chip tomorrow. So it's very exciting from a multimodal point of view for robotics assistance.
Finally, for creativity, we're the first model in the world—Gemini 2.0, which you can try now in AI Studio—that allows native image generation. Not calling a separate program or a separate model like Imagen 3, but Gemini itself natively generating images in the chat flow. People seem to be really enjoying using that. You're now talking to a multimodal chatbot. You can get it to express emotions in pictures, or give it a picture and tell it to modify it, and continue to work on it with word descriptions. "Can you remove that background? Can you do this?" This goes back to the earlier thing about programming or creative things in a new workflow. I think we're just seeing a glimpse of that if you try out this new Gemini 2 experimental model, in image creation. That's just the beginning. It will work with video, coding, and all sorts of things.
Reid Hoffman
In the land of the real world multimodal, one of the things people frequently speculate about is the geolocation of AI work. In the US, we intensely track everything happening on the West Coast. We also intensely track DeepMind, and then somewhat less, Mistral and others. What's some of the stuff that's really key for the world to understand coming out of Europe? What's the benefit of having multiple major centers of innovation and invention, not just within the West Coast, but also DeepMind in London, Mistral in Paris, and others? What are some things for people to pay attention to, why it's important, and what's happening within the UK and European AI ecosystem?
Demis Hassabis
We started DeepMind in London and are still headquartered here for several reasons. This is where I grew up, what I know, where I had all my contacts. But the competitive reasons were that we felt the talent coming out of universities in the UK and Europe was the equivalent of the top US ones. Cambridge, Imperial, and Oxford are up there with MITs, Harvards, and the Ivy Leagues. They're always in the top 10 on the university world tables. But in 2010, if you had a PhD in physics from Cambridge and you didn't want to work in finance at a hedge fund in the City, but wanted to stay in the UK and be intellectually challenged, there weren't that many options. There weren't that many deep tech startups. We were the first to prove that could be done. We were a big draw for the whole of Europe. We got the best people from technical universities in Munich, Switzerland, and so on. For a long while, that was a huge competitive advantage. Salaries were cheaper here than on the West Coast, and you weren't competing against the big incumbents.
The other reason I chose to do that was I knew the AGI plan from the beginning: solve intelligence and then use it to solve everything else. That was our mission statement. I still like that framing. It was a 20-year mission. We're now 15 years in, and I think we're on track, unbelievably, which is strange for any 20-year mission. But you don't want to be too distracted on the way in a deep technology, deep scientific mission. One of the issues I find with Silicon Valley is that it has lots of benefits—contacts, support systems, funding, amazing things, and the density of talent—but it is quite distracting.
Everyone and their dog is trying to do a startup they think is going to change the world, but it's just a photo app or something. The cafes are filled with this. Of course, it leads to some great things, but it's also a lot of noise if one wants to commit to a long-term mission that you think is the most important thing ever. You don't want your staff to be too distracted, thinking, "Maybe I could make a million if I jumped and did this gaming app quickly." That's the milieu you're in in the Valley, at least back then. Maybe this is less true now with more mission-focused startups. But I also wanted to prove it could be done elsewhere.
The final reason I think it's important is that AI is going to affect the whole world. It's going to affect every industry, every country. It's going to be the most transformative technology ever, in my opinion. If that's true, and it's going to be like electricity or fire—more impactful than even the internet or mobile—then I think it's important that the whole world participates in its design, with the different value systems and philosophies that are out there. From democratic values in Western Europe and the US, I think it's important that it's not just a hundred square miles of a patch of California. I do think it's important that we get these other inputs—broader inputs, not just geographically, but also from different subjects: philosophy, social sciences, economists, academia, civil society. Not just the tech companies, not just the scientists involved in deciding how this gets built and what it gets used for. I've always felt that very strongly from the beginning. Having some European involvement and some UK involvement at the top table of innovation is a good thing.
Aria Finger
Demis, one of the areas of AI that when anyone asks me, "Hey, Aria, I know you're interested in AI, but like, well, you can write my emails. Why is it so special?" I just say, "No, think about what it can do in medicine." I always talk about AlphaFold. I tell them about what Reid is doing. I'm just so excited for those breakthroughs. Can you give us just a little bit? You had this seminal breakthrough in AlphaFold. What is it going to do for the future of medicine?
Demis Hassabis
I've always felt that the most important things AI can be used for are two. One is human health—that's number one, trying to solve and cure terrible diseases. Number two is to help with energy, sustainability, and climate—the planet's health. Those are the two areas we have focused on in our science group, which I think is fairly unique amongst the AI labs, in terms of how much we push that from the beginning. Protein folding specifically was canonical for me. I came across it when I was an undergrad at Cambridge 30 years ago, and it's always stuck with me as this fantastic puzzle that would unlock so many possibilities.
The structure of proteins: everything in life depends on proteins, and we need to understand the structure so we know their function. If we know the function, then we can understand what goes wrong in disease, and we can design drugs and molecules that will bind to the right part of the surface of the protein if you know the 3D structure. It's a fascinating problem. It goes to all the computational things we were discussing earlier. Can you enumerate through this forest of possibilities? Some people estimate that Levinthal famously in the 1960s estimated an average protein can fold in 10 to the 300 possible ways.
How do you enumerate those astronomical possibilities? Yet it is possible with these learning systems. That's what we did with AlphaFold.
We spun out a company, Isomorphic, and I know Reid is very interested in this area too. If we can reduce the time it takes to discover a protein structure—it used to take a PhD student their entire PhD, four or five years, to discover one protein structure. There are 200 million proteins known to science, and we folded them all in one year. That's a billion years of PhD time in one year. We gave it to the world freely to use, and 2 million researchers around the world have used it. We spun out Isomorphic to go further downstream and develop the drugs needed, trying to reduce that time.
Aria Finger
It's just amazing. Demis, there's a reason they gave you the Nobel Prize. Thank you so much for all of your work in this area. It's truly amazing. Now to rapid fire. Is there a movie, song, or book that fills you with optimism for the future?
Demis Hassabis
There are lots of movies I've watched that have been super inspiring for me. Even something like *Blade Runner* is probably my favorite sci-fi movie, but maybe it's not that optimistic. If you want an optimistic thing, I would say the Culture series by Iain Banks. I think that's the best depiction of a post-AGI universe where you have AIs and societies of AIs and humans and alien species, with maximum human flourishing across the galaxy. That's an amazing, compelling future that I would hope for humanity.
Aria Finger
What is a question that you wish people asked you more often?
Demis Hassabis
The questions I sort of wonder why people don't discuss a lot more, including with me, are some of the really fundamental properties of reality that actually drove me in the beginning when I was a kid to think about building AI, to help us as an ultimate tool for science. For example, I don't understand why people don't worry more about what is time, what is gravity, or the fundamental fabric of reality. These are things staring us in the face all the time, very obvious things that impact us all the time, and we don't really have any idea how they work. I don't know why that doesn't trouble people more. It troubles me. I'd love to have more debates with people about those things, but most people seem to shy away from those topics.
Aria Finger
Where do you see progress or momentum outside of your industry that inspires you?
Demis Hassabis
That's a tough one because AI is so general. It's almost touching every industry. I'm not sure there are many outside of the AI industry. Maybe the progress going on in quantum is kind of interesting. I still believe AI is going to get built first and then will maybe help us perfect our quantum systems. I have ongoing bets with some of my quantum friends, like Hartmut Neven, on whether they're going to build quantum systems first and then that will help us accelerate AI. So I always keep a close eye on the advances going on with quantum computing systems.
Aria Finger
Final question. Can you leave us with a final thought on what is possible over the next 15 years if everything breaks humanity's way, and what's the first step to get there?
Demis Hassabis
What I hope for in the next 10 to 15 years is that what we're doing in medicine will really have new breakthroughs. I think in the next 10 to 15 years, we can actually have a real crack at solving all disease. That's the mission of Isomorphic. With AlphaFold, we showed the potential to do what I like to call "science at digital speed." Why couldn't that also be applied to finding medicines? My hope is that in 10 to 15 years, we'll look back on the medicine we have today a bit like how we look back on medieval times and how we used to do medicine then. That would be the most incredible benefit we could imagine from AI.
End of interview. Podcast outro and sponsor messages follow.