Transcription
We effectively, you can think of as 50% of our effort is on scaling, 50% of it is on innovation. My betting is you're going to need both to get to AGI. I've always felt this, that if we build AGI and then use that as a simulation of the mind and then compare that to the real mind, we will then see what the differences are and, uh, potentially what's special, um, and remaining about the human mind, right? Maybe that's creativity, maybe it's emotions, maybe it's dreaming. There's a lot of consciousness. There's a lot of, um, hypotheses out there about what may or may not be computable. And this comes back to the Turing machine question of like, what is the limit of a Turing machine?
>> So there's nothing that cannot be done within the sort of computational.
>> Well, no one's put it this way. Nobody's found anything in the universe that's that's non-computable, >> so far. >> So far.
Welcome to Google DeepMind, the podcast with me, Professor Hannah Fry. It has been an extraordinary year for AI. We have seen the center of gravity shift from large language models to agentic AI. We've seen AI accelerate drug discovery and multimodal models integrated into robotics and driverless cars. Now, these are all topics that we've explored in detail on this podcast. But for the final episode of this year, we wanted to take a broader view, something beyond the headlines and product launches to consider a much bigger question. Where is all this heading, really? What are the scientific and technological questions that will define the next phase? And someone who spends quite a lot of their time thinking about that is Demis, CEO and co-founder of Google DeepMind. Welcome back to the podcast, Demis.
>> Great to be back.
>> I mean, quite a lot's happened in the last year.
>> Yes. What's, what sort of the biggest shift do you think?
>> Oh wow. I mean, um, it's just so much has happened, as you said, it's just, it feels like we've packed in 10 years in one year. I think a lot's happened. I mean, certainly for us, uh, the, the progress of the models, um, we've just released Gemini 3, which we're really happy with. Um, the, the cap multimodal capabilities, all of those things have just advanced really well. And then probably the thing I, I guess over the summer that I'm, uh, very excited about is world models being advanced. I'm sure we're going to talk about that.
>> Yeah, absolutely. We will get on to all of that stuff in a bit more detail in a moment. Um, I remember the very first time that I interviewed you for this podcast and you were talking about the root node problems, about this idea that you can use AI to kind of unlock these downstream benefits. Um, and you've made pretty good on your promise, I have to say. And do you want to give us an update on on where we are with those, where, what are the things that are just around the corner and, and the things that we've, that you've sort of solved or near-solved?
>> Yeah. Well, of course, obviously the big proof point was was AlphaFold and, sort of crazy to think we're coming up to like, 5-year sort of anniversary of of AlphaFold being sort of announced to the world, AlphaFold 2 at least. So that was the proof, I guess, that it was possible to do these root node type of problems. And we're looking, we're exploring all the other ones now. I think material science, uh, I'd love to do a room temperature superconductor, um, and, uh, you know, better batteries, these kinds of things. I think that's that's on the cards, uh, better materials of all sorts. We're also working on fusion, um,
>> because there's a new partnership that's been announced, fusion.
>> Yeah, we've just announced partnership with a deep one. We, we already were collaborating with them, but it's a much deeper one now with Commonwealth Fusion, who, you know, I think are probably the best startup, uh, uh, working on at least traditional TOKAC, uh, reactors. So they're probably closest to to having something, uh, uh, viable, and we want to help accelerate that, uh, you know, helping them contain the plasma in the magnets and maybe even some material design there as well. So that's exciting. And then we're collaborating also with our quantum colleagues, which they're doing amazing work, uh, at the at the quantum AI team at Google, and we're helping them with error correction codes, uh, where we're using our machine learning to help them, and then maybe one day they'll help us.
>> That's perfect. Exactly. The fusion one is particularly, I mean, the difference that that would make to the world, that would be unlocked by that is gigantic.
>> Yeah. I mean, fusion's always been the holy grail. Of course, I think solar is very promising too, right? Effectively using the fusion, fusion reactor in the, in the, in the clouds in the sky. But, um, I think if we could have, uh, modular fusion reactors, you know, this promise of, uh, almost unlimited renewable clean, uh, energy, uh, would be obviously transform everything. And that's the holy grail. Of course, that's one of the ways we could we could, um, uh, help with climate.
>> Does make a lot of our existing problems sort of disappear if we can, if we can.
>> Definitely. I mean, it opens up many. This is why we think of as a root node. Of course, it helps directly with energy and and pollution and and so on, um, and helps with the, with the climate crisis. But also, if energy really was renewable and clean and and and super cheap or almost free, then many other things would become viable. Um, like, you know, water access, 'cause we could have desalination plants pretty much everywhere. Uh, even making rocket fuel. Uh, you know, it's just there's lots of seawater that contains hydrogen and oxygen. That's basically rocket fuel, but it just takes a lot of energy to split it out into hydrogen and oxygen. But if energy is cheap, uh, and and renewable and sort of clean, then why not do that? you know, you could have that producing 24/7.
>> You're also seeing a lot of change in the, uh, the AI that is applying itself to mathematics, right? That, you know, winning medals in the International Math Olympiad and yet at the same time, these models can make quite basic mistakes in high school math. Why is there that paradox?
>> Yeah, I think it's fascinating, actually. One of the most fascinating things and probably that needs to be fixed, uh, as one of the key things while we're not at AGI yet. Um, as you said, we've had a lot of success in other groups on getting like gold medals at the International Math Olympiad. You look at those questions and they're, they're super hard questions that only the top students in the world can, can do. And on the other hand, if you pose a question in a certain way, we've all seen that with with experimenting with chatbots ourselves, uh, in our daily lives, that it can make some fairly trivial mistakes on logic problems. They can't really play decent games of chess yet, which, um, is surprising. So there's something missing, um, uh, still from these systems in terms of their consistency. And I think that's one of the things that's, you would expect from a general intelligence and, art, you know, an AGI system, is that it would be consistent across the board. And so sometimes people call it jagged intelligences. So they're really good at certain things, maybe even like PhD level, but then other things they're like, not even high school level. So it's very uneven still, the performances of these systems. They're very, very impressive in certain dimensions, um, but they're still pretty, uh, uh, basic in others, and we've got to close those gaps. And, you know, there are theories, reasons to why, and depending on the situation, it could even be, uh, the way that an image is, is perceived and tokenized. So sometimes, actually, it doesn't even get all the letters that you, you know, so when you count letters in words, um, it sometimes gets that wrong, but, but it may not be seeing that each individual letter. So there's sort of different reasons for some of these things. Uh, and each one of those can be fixed, and then you can see what's left. Um, but I think consistency, I think another thing is reasoning and thinking. So, uh, we have thinking systems now that at inference time, they spend more time thinking, and they're better, they're better at outputting their answers. Um, but it's not sort of super consistent yet in terms of like, is it using that thinking time in a useful way, um, to actually double-check and use tools to double-check what it's outputting? I think we're, we're on the way, but maybe we're only 50% of the way there.
>> I also wonder about that, that story of of AlphaGo and then AlphaZero, where you sort of took away all of the human experience and found that the model actually improved. Is there a sort of, is there a scientific or a math version of that in in the models that you're creating?
>> I think what we're trying to build today, it's more like AlphaGo. So, you know, you effectively, these, these large language models, these foundation models, they're starting with all of human knowledge, you know, what we put on the internet, which is pretty much everything these days, and, um, compressing that into some useful artifact, right, which they can look up and, and generalize from. But I do think we're, we're still in the, in the early days of having this, uh, uh, search or thinking on top, like AlphaGo had to kind of, uh, use that model to direct in useful reasoning traces, useful planning, uh, ideas. Um, and, and then come up with the best, you know, solution to whatever the problem is at that point in time. So I, I don't feel like we're constrained at the moment with the kind of limit of human knowledge, like the internet. I think the main issue at the moment is we don't know how to use those systems in a reliable way fully yet, in the way we did with AlphaGo. Um, but of course, that was a lot easier because it was just, it was a game. I think once you have AlphaGo there, uh, you could go back, just like we did with the, the Alpha series, and do an AlphaZero where it starts sort of discovering knowledge for itself. I think that would be the next step, but I, I, that's obviously harder, and so I think it's good to try and create the first step first with some kind of Alpha-like system, and then we can think about an AlphaZero-like system. But that is also one of the things missing from today's systems is the ability to online learn and continually learn. So, you know, we train these systems, we balance them, we post-train them, and then they're out in the world, but they don't do, they don't continue to learn out in the world like, like we would. Um, and I think that's another missing, critical missing piece from, from these systems, from, from, you know, that will be needed before AGI.
>> In terms of all of those missing pieces, I mean, I know that there's this big race at the moment to release commercial products, but, but I also know that that Google DeepMind's roots really lie in in that idea of scientific research. And I, I found a quote from you where you recently said, "If I had had my way, we would have left AI in the lab for longer and done more things like AlphaFold, maybe cured cancer or something like that." Mhm.
>> Do, do you think that we lost something by not taking that slower route?
>> Um, I think we lost and gained something. So I feel like that would have been the more pure scientific approach. At least that was my original plan say 15, 20 years ago, that, you know, when almost no one was working on AI, we just started, we were just about to start DeepMind. People thought it was a crazy thing to work on. Um, but we believed in it, and, and, and, and I think that the idea was, if we would make progress, we would continue to sort of, um, incrementally build towards AGI, be very careful about what each step was, and, and the safety aspects of it, and so on, analyze what the system was doing, and so on. But in the meantime, you wouldn't have to wait till AGI arrived before it was useful. You could branch off that technology and use it in really beneficial ways to society, namely advancing science and medicine. So, exactly what we did with AlphaFold, actually, which, um, it's not, it's not a foundation model itself, general model, but it uses the same techniques, you know, transformers and other things, and then blends it with, um, uh, more specific things to that domain. So I imagined a whole bunch of those things getting done while, which would be hugely, you know, you'd release to the world for, just like we did with AlphaFold, and indeed do things like cure cancer and so on, um, whilst we were working on the sort of more the AGI track in the lab. Now, it's turned out, uh, that chatbots were possible at scale and people find them useful, and then they've now morphed into these foundation models that can do more than chat and text, obviously, including Gemini. Uh, they can do images and video and all sorts of things, and, um, that's also been very successful commercially and in terms of a product, and I love that too. Like, I've always dreamed of having the ultimate assistant that would help you in everyday life, make it more productive, maybe even protect your brain space a bit as well from an attention, so that you can focus and be in flow and so on, 'cause, you know, today with social media, it's just noise, noise, and I think AI can actually, that works for you, could help us with that. Um, so I think that's good, but it has created this pretty crazy race condition where there's many commercial organizations and even nation states all rushing to, you know, improve and overtake each other, and that makes it hard, uh, to do sort of rigorous science at the same time. We try to do both, and I think we're getting that balance right. On the other hand, there are lots of pros of the way it's happened, which is, of course, there's a lot more resources coming, uh, into the area. So that's definitely accelerated progress. Um, and also, um, I think the general public are actually, interestingly, only a couple of months behind the absolute frontier in terms of what they can use. So everyone gets the chance to sort of feel for themselves what AI is going to be like. And I think, I think that's a good thing, and then governments, sort of understanding this better.
The thing that's strange is that, I mean, this time last year, I think there was a lot of talk about, you know, scaling eventually hitting a wall, about us running out of data. And yet, you know, we're recording now, Gemini 3 has just been released, and it's leading on this whole range of different benchmarks. Um, how, how has that been possible? Like, wasn't there supposed to be a problem with scaling hitting a wall?
I think a lot of people thought that, especially as other companies have sort of had slower progress, should we say. But I think we've never really seen any wall as such. Like, what I would say is, um, maybe there's like diminishing returns. And people, when I say that, people think only think like, oh, so there's no returns, like it's zero or one, it's either exponential or or it's asymptotic. No, actually, there's a lot of room between those two regimes, and I think we're in, in between those. So it's not like you're going to double the performance on all the benchmarks every time you release a new iteration. Maybe that's what was happening in the early, very early days, you know, three, four years ago. But you are getting significant improvements, like we've seen with Gemini 3, that are well worth the investment and the return on that investment and doing. So, I, that we haven't seen any slowdown on. There are issues like, are we running out of just available data? But there are ways to get around that, you know, synthetic data, generating your, you know, these systems are good enough, they can start generating their own data, especially in certain domains like coding and math, where you can verify the answer in some sense, you could produce unlimited data. So all of these things, though, are research questions. And I think that's the advantage that we've always had is that, um, we've, we've always been sort of research first. And we, I think we have the broadest and deepest research bench, always have done. Um, and if you look back at the last decade of advances, whether that's transformers or AlphaZero, any of the things we just discussed, that they all came out of Google or DeepMind. So I've always said, like, if, if more innovations are needed, scientific ones, then I would back us to be the place to do it, just like we were, you know, in the previous sort of 15 years for a lot of the big breakthroughs. So I think that's just what's transpiring. And I actually really like it when the terrain gets harder, because then it's not just world-class engineering you need, which is already hard enough, um, but you have to ally that with world-class research and science, which is what we specialize in. Um, and on top of that, we also have the advantage of world-class infrastructure with our TPUs and, and other things that we've invested in a lot for a long time. Um, and so that combination, I think, allows us to, uh, uh, sort of be at the frontier of the innovations, as well as the scaling part. And we effectively, you can think of as 50, 50% of effort is on scaling, 50% of it is on innovation. And I think my betting is you're going to need both to get to AGI.
>> I mean, one thing that we are still seeing even in Gemini 3, which is an exceptional model, is, uh, this idea of hallucinations. So I think, um, there was one metric that said, uh, it can still give an answer when actually it should decline. Um, I mean, could you build the system where Gemini gives a confidence score in the same way that AlphaFold does?
>> Yeah, I think so. And I think we need that, actually. And I think that's sort of one of the missing things. I think the better the models get, the more they know about what they know, if that makes sense. And so, and I think the more reliable we could sort of rely on them to actually introspect in some way or do more thinking and actually realize for themselves that they're uncertain or there's, there's, there's uncertainty over this answer. Uh, and then we've got to sort of work out how to train it in a way that where it can, it can output that as a, as a reasonable answer. Um, we're getting better at it, but it still sometimes, you know, it sort of forces itself to answer when it probably shouldn't. Um, and then that can lead to a hallucination. So, I think, you know, a lot of the hallucinations are of that type currently. So, there's a missing piece there that that sort of has to be solved. And you're right, as we did solve it with AlphaFold, but in, in obviously a much more limited way.
>> Cuz presumably behind the scenes, there is some sort of measure of probability of whatever the next token might be.
>> Yes, there is of the next token. That's how it all works. But that doesn't tell you the overall arching piece is this is, you know, how confident are you about this entire fact or this entire, um, statement. And I think that's why you'll need this. I, I think we'll need to use the thinking steps and the planning steps to go back over what you just output. At the moment, it's a little bit like the systems are just, it's like talking to some, a person, and they just, you know, when, when they're in on a bad day, they're just literally telling you the first thing that comes to their mind. Most, most of the time, that would be okay. But then sometimes when it's a very difficult thing, uh, you'd want to like stop, pause for a moment, and maybe go over what you were about to say and adjust what you were about to say. But perhaps that's happening less and less in the world these days, but, um, that's still the better way of having a discourse. So, you know, I think you can think of it like that. These models need to do that better.
>> I also really want to talk to you about, um, the, the simulated worlds and putting agents in them, because we got to talk to your Genie team earlier today.
>> Tell me why you care about simulation. What, what can a world model do that, that a language model can't?
>> Well, look, I, it's, it's actually been, it's probably my longest-standing passion is world models and simulations, uh, in addition to AI. And of course, it's all coming together in our most recent work, like Genie. And I think, um, language models are able to understand a lot about the world. I think actually more than we expected, more than I expected, because language is actually probably richer than we thought. It contains more about the world than we maybe even, even linguists, maybe imagined. And that's, you know, proven now with these new systems. But there's still a lot about the, the spatial dynamics of the world, you know, how spatial awareness, um, and the cont, the physical context we're in, um, and how that works mechanically, that, um, isn't, is hard to describe in words and isn't generally described in in corpuses of of words. And a lot of this is allied to learning from experience, online experience. There's a lot of things which you can't really describe, something you have to just experience it. Um, maybe the sensors and so on are very hard to put into words, you know, whether that's, you know, motor angles and smell and, you know, these kind of sensors, it's very difficult to describe that in any kind of language. So I think there's a whole set of things around that. And I think if we want robotics to work, or a universal assistant that maybe comes along with you in your daily life, maybe on glasses or, you know, on your phone, um, and helps you in your everyday life, not just on your computer, um, you're going to need this kind of world understanding. And, uh, world models are at the core of that. So this, what we mean by a world model, is this, this sort of model that understands the causitative and effect of of the mechanics of the world, right? Intuitive physics. But, um, how things move, how things behave. Um, now, we're seeing a lot of that in our video models, actually. And one way to show, how do you test, have that kind of understanding? Well, can you generate realistic worlds? 'Cause if you can generate it, then in a sense, you must have understood, uh, uh, the system must have encapsulated a lot of the mechanics of the world. So that's why Genie and VEO, and these models are our video models, and our sort of interactive world models are really, uh, impressive, but also important steps towards showing we have generalized models. And then hopefully, at some point, we can apply it to, you know, robotics and, and, and universal assistance. And then, of course, one of my favorite things, I'm definitely going to have to do at some point, is re-applying it back to games and, and, you know, game simulations and create the ultimate games, which of course was maybe always my subconscious plan.
>> All of this.
>> Yeah. All of the time. Exactly.
>> What about science too, though? Could you use it in that, in that domain?
>> Yes, you could. So, uh, science, you know, again, I think building models of scientifically complex, uh, uh, domains, uh, whether that's materials on an atomic level, um, you know, in biology, uh, but also like some physical things as well, like weather. One way to, um, understand those systems is to build simula, learn simulations of those systems from the raw data, right? So you have a bunch of raw data, let's say it's about the weather, and obviously we have some amazing weather projects going on, um, and then you have a model that kind of learns those dynamics and can recreate those dynamics, uh, more efficiently. So, uh, than doing it by brute force. So I think there's huge potential for simulations and, uh, kind of world models, maybe specialized ones for aspects of of science and mathematics.
>> But then also, I mean, you can drop an agent into that simulated world too, right?
>> Yes. Your Genie 3 team, they had this really lovely quote, which was, "Almost no prerequisite to any major invention was made with that invention in mind." And they were talking about dropping agents into these simulated environments and allowing them to explore with sort of curiosity being their main motivator.
>> Right. And so that's, that's another really exciting use of these, these, uh, world models. Is you can, we have another project called Simma. We just, we just released Simma 2. Sim, you know, simulated agents, where you have an avatar or an agent, and you put it down into a virtual world. It can be a normal, it can be a kind of actual commercial game or something like that, very complex one, like No Man's Sky, kind of open-world space game. Uh, and then you can, you can instruct it with, because it's got Gemini under the hood, you can just talk to the agent and, and give it, give it tasks. But then we thought, well, wouldn't it be fun if we plugged Genie into Simma and sort of dropped Simma, a Simma agent into another AI that was creating the world on the fly? So now the, the two AIs are kind of interacting in the minds of each other. So Simma's, you know, the Simma agent's trying to navigate this world, and Genie is, as far as Genie is concerned, that's just a player and an avatar, doesn't care. There's another AI. So it's just generating the world around whatever Simma is trying to do. So, so it's kind of amazing to see them both, uh, interacting together. And I think this could be the beginning of an interesting training loop, where, uh, you almost have infinite training, uh, examples, because, uh, whatever the Simma agent's trying to learn, Genie can basically create on the fly. So, I think you could imagine a whole world of like, uh, setting and solving tasks, just millions of tasks automatically, and they're just getting increasingly more difficult. So, we might try to set up a kind of loop like that, um, as well as, obviously, those Simma agents could be great as game companions. Um, also some of the things that they learn could be useful also for robotics.
>> Yeah. The end of boring NPCs, basically.
>> Exactly. It's going to be amazing for these games. Yeah.
>> Those worlds that you're creating, though, how do you make sure that they really are realistic? I mean, how do you ensure that you don't end up with physics that looks plausible, but is actually wrong?
>> Yeah, it, that's, that's a great question, and, and, and can be an issue. It's basically hallucinations again. So some hallucinations are good, 'cause, 'cause you, it also means you, you might create something interesting and new. So, in fact, sometimes if you're trying to do, create creative things or trying to get your system to create new things, novel things, um, a bit of hallucination might be good, but you want it to be intentional, right? So not, uh, so you kind of switch on the hallucinations now, right? Or the, the creative, um, exploration. But yes, with, with the, when you're trying to train a Simma agent, you don't want Genie hallucinating, kind of physics that are wrong. So, actually, what we're doing now is we're almost creating a physics benchmark, where, um, we can use game engines, which are very accurate with physics, to create lots of like, um, fairly simple, like the sorts of things you would do in your physics A-level, uh, lab, uh, lessons, right? Like, you know, rolling little balls down different tracks and seeing how fast they go and so, like, really teasing apart on a very basic, uh, level, like Newton's three laws of motion, has it encapsulated it, um, whether that's VEO or Genie, have these models encapsulated the physics of that 100% accurately? And right now, they're not. They're kind of approximations, and they look, um, realistic when you just casually look at them, but they're not, uh, they're not accurate enough yet to rely on for say, robotics. So now we've got these really interesting models. Um, and with physics, I think that's going to probably involve generating loads and loads of ground truth, simple videos of pendulums, you know, what happens when two pendulums go around each other, but then very quickly you get to like, three-body problems, which are not solvable anyway. So I think it's going to be interesting. But what's amazing already is when you look at the, the video models like VEO, and just the way it treats reflections and liquids, it's pretty unbelievably accurate already, at least to the naked eye. So the next step is actually going beyond what a human can, amateur can perceive, and, would it really hold up to a proper physics-grade experiment?
I know you've been thinking about these simulated worlds for a really long time, and, uh, I went back to the transcript of our first interview, and in it, you said that you really like the theory that consciousness was this consequence of evolution, um, that, you know, at some point in our evolutionary past, there was like an advantage to understanding the internal state of another, and then we sort of turned it in on ourselves.
Does that make you curious about running, sort of, an agent in evolution inside of a simulation?
Sure. Um, I mean, I'd love to run that experiment at some point. Kind of rerun evolution, rerun, um, almost social dynamics as well. Like, the Santa Fe used to run lots of cool experiments on little grid worlds. I used to love some of these, but they're mostly economists, and they were trying to like, you know, run like little, uh, artificial societies, and they found that things, all sorts of interesting things got invented, like that, uh, if you let agents run around for long enough with the right incentive structures, markets and banks and all sorts of crazy things. So I think it would be really cool, and also just to understand the origin of life and the origin of consciousness. And I think that is the one of the big passions I had for for working on AI from the beginning was, I think you're going to need these kinds of tools to really understand where we came from and what these phenomena are. Um, and I think simulations is, is, is one of the most powerful tools to do that, because you can then do it statistically, because you can run the simulation many times with control, slightly different initial starting conditions, and then, um, maybe run it millions of times, and then understand what the slight differences are in a very, uh, controlled experiment sort of way, which, of course, is, you know, very difficult to do in the real world for any of the really interesting questions we want to answer. So I think accurate simulations will be an unbelievable boon to science.
>> Given, you know, what we've discovered about sort of emergent properties of these models, right, having sort of conceptual understanding that we weren't expecting, do you also have to be quite careful about running this sort of simulation?
>> Uh, I think you would have to be, yes. But that, that's the other nice thing about simulations, you can run them in, you know, pretty safe sandboxes, maybe eventually you want to air-gap them, uh, and you can, of course, monitor what's happening in the, in the, in the simulation 24/7, uh, and you have access to all the data. So we may need AI tools to help us monitor the simulations, because, um, they'll be so complex, they'll be, and there'll be so much going on in them. If you imagine loads of AIs running around in a simulation, uh, uh, it will be hard for any human scientist to keep up with it, but we could probably use other AI systems to help us analyze and flag anything interesting or worrying in those simulations, uh, automatically. I mean, this, I guess we're still talking sort of medium to long-term in terms of this stuff. So, so just going back to the trajectory that we're on at the moment. Um, I also want to talk to you about the, the impact that AI and AGI are going to have on, on wider society.
>> Um, and last time we spoke, you said that you thought AI was overhyped in the short term, but underhyped in the long term.
>> Um, and I know that this year, there's been a lot of chatter about an AI bubble.
>> Yes.
>> What happens if there is a bubble and it bursts? What happens?
Well, look, I, I think, yes, that I still subscribe to it's overhyped in the short term, still, and still underappreciated in the, in the medium to long term, what's going to, you know, how transformative it's going to be. Um, yeah, there is a lot of talk, of course, right now about AI bubbles. Um, in my view, uh, I, I think it, there isn't, it's not one thing, binary thing, are we or aren't we? I think there are parts of the AI ecosystem that are probably in bubbles. What, one example would be, you know, just seed rounds for startups, uh, that basically haven't even got going yet, and they're raising at tens of billions of dollars, uh, valuations just out of the gate. It's sort of interesting to see how, how can that be sustainable? Um, you know, my guess is probably not, uh, at least not in general. Um, so there's that area. Then the people are worrying about, obviously, there's, there's the big tech valuations and other things. I think there's a lot of real business underlying that. So, um, but it remains to be seen. I mean, I think maybe for any, any, uh, new, unbelievably transformative and profound technology, of which, of course, AI is probably the most profound, uh, you're going to get this, uh, overcorrection in a way. So when we started DeepMind, no one believed in it. No one thought it was possible. People were wondering, what's AI for anyway? And then now, fast forward 10, 15 years, and now, obviously, it seems to be the only thing people talk about in business. And, um, so it's a, but you're sort of going to get, it's almost an overreaction to the underreaction. Um, so I think that's natural. I think we saw that with the internet. I think we saw with mobile, and I think we're, we're seeing or going to see it again with AI. Um, I don't worry too much about, are we in a bubble or not, because from my perspective, as you know, leading Google DeepMind, and also obviously with Google as, and Alphabet as a whole, our job and my job is to make sure either way, we, uh, come out of it very strong. And I think, and we're very well positioned, and I think we are tremendously well-positioned either way. So if it continues going like it is now, fantastic. We'll carry on, you know, all of these great things that we're doing and experiments and progress towards AGI. If there's a retrenchment, fine. Then also, I think we're in a great position because, uh, we have our own stack with TPUs. We also have, um, all these incredible Google products, and, you know, the profits that all makes to plug in our AI into. And we're doing that with Search, is totally revolutionized by AI overviews, AI mode, with Gemini under the hood. We're looking at Workspace, at email, you know, at YouTube. So, there's all these amazing things in Chrome. There's a lot of these amazing things that, um, AI, we can see already are low-hanging fruit to apply, uh, Gemini 2, as well, of course, as Gemini app, which is doing really well as well now, and, and, and the idea of universal assistant. So, there's new products, and I think they will, in the fullness of time, be super valuable, but we don't have to rely on that. We can just power up our existing, uh, ecosystem. Uh, which is all sort of, I think that's what's happened over the last year. We've got that really efficient now.
In terms of the, the AI that people have access to at the moment, I, I know you said recently how important it is not to build AI to maximize user engagement, just so we don't repeat the, the mistakes of social media. But, but I also wonder whether we are already seeing this in a way. I mean, people spending so much time talking to their chatbots that they end up kind of spiraling into self-radicalizing. Yeah.
Um, how do you stop that? How do you build AI that, that puts users at the center of their own universe, which is sort of the point of this in a lot of ways, but without creating echo chambers of one?
Yeah, it's a very, you know, um, careful balance that, you know, I think is one of the most important things that we as an industry have got to get right. So I think we've seen what happens with, uh, you know, some systems that were overly sycophantic, or, you know, then you get these, these sort of echo chamber reinforcements that are really bad for the person. So I think part of it is, and actually what we want to build with, with Gemini, and I'm really pleased with the Gemini 3 persona that we had a great team working on, and I helped with too, personally, is, um, just this sort of almost like a scientific, uh, personality that's, um, it's warm, it's helpful, it's light, but it's, it's, it's succinct to the point, and it will push back on things in a friendly way that don't make sense. You know, rather than trying to reinforce you, you know, the idea that the earth's flat, and you said it, and it's like, wonderful idea, you know, I don't think that's good in general for society if that were to happen. Um, but you've got to balance it with what people want, 'cause people want, uh, these systems to be supportive, um, to be helpful with their, with with their ideas and their brainstorming. So, you've got to get that balance right. And I think, I think we are, we're sort of developing a science of of personality and persona, of like how to, to, to kind of measure what it's doing and where do we want it to be, like on authenticity, on humor, you know, these sorts of things. And then you can imagine there's a kind of base personality that it ships with. And then everyone has their own preferences. You know, do you want it to be more humorous, less humorous, or, or more succinct, or more verbose? People like different things. So you add that additional personalization layer on it as well. But there's still the core base personality that everyone gets, right? Which is trying to, trying to adhere to the scientific method, which is the whole point of these. And we want people to use these for science and for medicine and health issues and so on. Uh, and so, um, I think it's, it's, it's part of the science of getting these, uh, large language models right. And, um, I'm, I'm quite happy with the direction we're going in currently.
Uh, we got to talk to Shane Le a couple weeks ago, um, about, uh, AGI in, in particular, across everything that's happening in, in AI at the moment, the language models, the world models, you know, and so on. What's closest to your vision of AGI?
I think actually the combination of, obviously, there's Gemini 3, which I think is very capable. But the Nano Banana Pro system, we also launched last week, which is an advanced version of our image creation tool. What's really amazing about that, it has also Gemini under the hood. So, it can understand not just images, it sort of understands, uh, what's going on semantically in those images. Um, and people have been only playing with it for a week now, but I've seen so much cool stuff on on social media about, uh, what people are using it for. So, for example, um, you know, you can give it a picture of a, of a, of a complex plane or something like that, and it can label all the diagrams of, uh, you know, all the different parts of the plane, and even visualize it in, in a, for like, with all the different parts sort of exposed. Um, so it has some kind of deep understanding of mechanics and, and what make, what, you know, makes up parts of objects, what's materials. So it's a sort of, um, and it can, you know, render text really, really, uh, accurately now. So I think that's sort of, um, it's getting towards a kind of AGI for imaging. Um, I think it's, uh, a kind of general-purpose system that can do anything across images. So I think that's very exciting. And then the advances in, in world models, you know, Genie and Simma, and what we're doing there. And then eventually, we've got to kind of converge all of those different, they're kind of different projects at the moment, and they're, they're, they're intertwined, but we need to, you know, converge them all into one, one big model, and then that might be start becoming, you know, candidate for proto-AGI.
I know you've been reading quite a lot about the industrial revolution recently. Um, are there things that we can learn from what happened there to try and mitigate against the, the sort of, some of the disruption that, that we can expect AGI to bring?
I think there's a lot we can learn. It's, it's something you sort of study in school, at least in the, in the, in Britain, but, but on a very superficial level. Like, it was really interesting for me to look into how it, how it all happened, what it started with, the reasons behind the economic reasons behind that, which is like the textile industry, and then the first computers were really the sewing machines, right? And then they became punch cards for the early FORTRAN computers, mainframes. And for a while, it was very successful. In Britain, became like the center of the, the textile world, because they could make these amazingly high-quality things for very cheap, uh, because of the automated systems. Um, and then obviously the steam engines and all of those things came in. I think there's a lot of, um, incredible advances that came out of the industrial revolution. So, um, child mortality went down, and all of modern medicine, and, and, um, sanitary conditions, the kind of work-life, uh, uh, split, and how that all worked was kind of worked out during the industrial revolution. But it also came with a lot of challenges, like, it took quite a long time, um, roughly a century, and, um, different parts of the labor force were dislocated at certain times, and then new, uh, things had to be created, new organizations, like unions and other things, had to be created in order to rebalance that. So, like, it was fascinating to see the whole of society, sort of had to, over time, adapt. And then you've got the modern world now. So there were, I think there were lots of obviously pros and cons of the industrial revolution, why it was happening, but no one would want, if you think about what it's done in total, like abundance of, you know, people, you know, of food, and in the Western world, and, and modern medicine, and all these things, modern transport, that was all because of the industrial revolution. So, we wouldn't want to go back to pre-industrial revolution, but maybe we can figure out ahead of time by learning from it, what those dislocations were, and maybe mitigate those, um, earlier or more effectively this time. And we're probably going to have to, because the difference this time is that it's probably going to be 10 times bigger than the industrial revolution, and it'll probably happen 10 times faster. So, more like a decade, then unfold over a decade than a century.
One of the things that Shane told us was that that the kind of current economic system where, you know, you exchange your labor for resources, effectively, it, it just won't function the same way in a post-AGI society. Do you have a vision of of how society should be reconfigured, or might be reconfigured, in a way that works?
Yeah, I'm spending more time thinking about this now, and Shane's actually leading an effort here on that, to sort of think about what a post-AGI world might look like and what we need to prepare for. But I think society in general needs to spend more time thinking about that. Economists and social scientists and governments, because I, I, as with the industrial revolution, you know, the whole working world and working week and everything got changed from from pre-industrial revolution, war, agriculture. And I think that's going to at
least that level of change is going to happen again. So it's not surprising. I don't would not be surprised if we needed new economic systems, new economic models to uh to basically um help with that transformation and make sure for example the benefits are widely um distributed and maybe things like universal basic income and things like that are part of the solution.
But I don't think that's the complete uh I think that's just what we can model out now, right? Because that would be a almost an add-on to what we have today. But I think there might be something way better systems where um more like direct democracy type systems where you can you know vote with a certain amount of of credits or something for what you want to see. It happens actually on local uh community level. You know here's a bunch of money. Do you want a playground or a tennis court or an extra classroom on the school? And then you let the community um sort of vote for it, right? So, and then and then maybe you could even measure the outcomes and then and then the people that sort of consistently vote for the for for things that that end up being um more well-received, they they have proportionally more influence for the next vote. So, there's there's a lot of interesting things I hear, you know, economist friends of mine who are are kind of brainstorming this and I think that would be great if we had a lot more work on that.
And then there's the philosophical side of it of like okay so jobs will change and other things like that but then um but maybe we'll have fusion will have been solved and so we have this sort of abundant free energy so we're post scarcity so what happens to money um maybe everyone's better off but then what happens to purpose right because a lot of people get their purpose from you know their jobs and then providing for their families uh which is a very noble purpose so if that's you know so there's a lot of I I think some of these questions blend from economic questions into almost philosophical questions.
>> Do you do you worry that people don't seem to be paying attention sort of or moving as quickly as you'd like to see? What would it take for for people to sort of recognize that we need international collaboration on this?
I am worried about that and I wish that and and again in a sort of ideal world there would have been a lot more collaboration already and international specifically uh and a lot more research and and sort of um I guess exploration and discussion going on about these topics. I'm actually pretty surprised there isn't more of that being discussed given that you know even our timelines which were there are some very short timelines out there but even ours are 5 to 10 years which is not long for for for for institutions or things like that to be built to to handle this. Um, and one of the worries I have is that the institutions that do exist they you know seem to be very fragmented and not very influential to to the level that you would need. Um, so it may be that that that that there are there aren't the right institutions to deal with this currently. And then of course if you add in the geopolitical tensions that are going on at the moment around the world, it seems like collaboration, cooperation is harder than ever. Um, just look at climate change and and um how hard it is to get any agreement on anything to do with that. So um so we'll see. I think as the stakes get higher and as these systems get more powerful and maybe this is one of the benefits of them being in products is uh the the you know everyday uh person that's not working on this technology will get to feel the increase in the power of these things and the capability and so that will then reach government and then maybe um uh they'll see sense as we get closer to to AGI.
>> Do you think it will take a moment an incident for everyone to sort of sit up and pay attention?
>> I don't know. I mean, I hope not. Most of the main labs are pretty pretty responsible. We try to be as responsible as possible. You know, that's always something we've, as you know, if you followed us over the years, that's been at the heart of what everything we do. Doesn't mean we'll get everything right, but we try to be as thoughtful and as scientific in our approach as possible. Um, I think most of the major labs are are trying to be responsible. Also, there's good commercial pressure actually to be responsible. If you think about agents, uh, and you're renting an agent to another company, let's say, to do something, um, that other company is going to want to know what the limits are and the boundaries are and the guardrails are on those agents, you know, in terms of what they might do and not just mess up the data and all of this stuff. So, I think that's good because the pe the more kind of carboy operations, they won't um get the business because the enterprises won't choose them. So I think the kind of capitalist system will actually be useful here to reinforce responsible behavior which is good but then there will be rogue actors um maybe rogue nations maybe rogue organizations um maybe people building on top of open source I don't know like obviously it's very difficult to stop that then um something may go wrong and uh hopefully it's just sort of medium-sized and then that will be a kind of warning shot to to to humanity across the bow and then that might be the moment to kind of um advocate for uh international uh standards or international cooperation or collaboration at least on some the high level basic or you know kind of like what's the basic standards we we we would want and and and agree to I'm hopeful that that will be possible
>> in the long term so beyond AGI and and towards ASI right artificial super intelligence do you think that there are some things that that humans can do that machines will ever be able to manage?
>> Well, I think that's the big question and I feel like this is related to as you know, one of my favorite topics is cheuring machines. I've always felt this that if we build a GI and then use that as a simulation of the mind and then compare that to the real mind, we will then see what the differences are and uh potentially what's special um and remaining about the human mind, right? Maybe that's creativity, maybe it's emotions, maybe it's dreaming. There's a lot of consciousness. There's a lot of um hypotheses out there about what may or may not be computable. And this comes back to the chewing machine question of like what is the limit of a chewing machine? And I think that's the central question of my life really ever since I found out about chewing and chewing machines. And um you know I think that's that's I fell in love with that. That's my core passion. And I think um everything we've been doing is been sort of pushing the notion of what a cheuring machine can do to the limit including you know folding proteins right and so it turns out I'm not sure what the limit is maybe there isn't one right and of course the my quantum computing friends would would say there are limits and and you need quantum computers to do quantum systems but I'm really not so sure and I've actually you know discussed that with some some some of the quantum folks and it may that we need data from these quantum systems in order to create a classical simulation. Um, and then that that comes back to the mind which is is it all classical computation or is there something else going on you know like Roger Penrose believes you know there's quantum effects in the brain. If there are then and that's what consciousness is do with then machines will never have that at least the the the classical machines we'll have to wait for quantum computers. Um, but if they if there isn't then there may not be any limit maybe in the universe everything is computationally tractable and therefore if you look at it in the right way and therefore chewing machines might be able to model everything in the universe I I'm currently if you were to get make me guess I would guess that and I'm working on that basis until physics um shows me otherwise
>> so there's nothing that cannot be done within these sort of computational
>> well no one's put it this way nobody's found anything in the universe that's that's non-computable So far
>> so far,
>> right? And I think we've already shown you can go way beyond the the usual complexity theorist P= MP view of like what a classical computer could do today. Things like protein folding and go and so on. So I don't think anyone knows what that limit is. And that's really if you boil down to what we're doing at Deep Mind and Google and what I'm trying to do is is find that limit. But then in the limit of that though, right, is that in the limit of that idea is that, you know, we're sitting here sort of there's like the warmth of the lights on our face. We kind of hear the wear of the machine in the background. There's like the feel of the desk under our hands.
>> All of that could be
>> replicable by a classical computer.
>> Yes. Well, I think in the end, my view on this is why I love K as well is all all of all all of my two favorite philosophy is a construct of the mind. I think that's true. And so, yes, all of those things you mentioned, they're coming into our sensory apparatus and they feel different, right? The light, the warmth of the light, the feel, the touch of the table, but in the end, they're it's all information. And we're information processing systems. And I think that's what biology is. This is what we're trying to do with isomeorphic. That's how I think we'll end up curing all diseases is by thinking about biology um as an information processing system. And I think in the end that's going to be and I'm working on my spare time, my 2 minutes of spare time, you know, physics theories about uh things like information being the most fundamental unit, should we say, of the universe, not energy, not matter, but information. And um so it may be that these are all interchangeable in the end, right? But we just sense it. We feel it in a different way. Um, but you know, as far as we know, this is still all these amazing sensors that we have, they're still computable by a chewing machine. But this is why your simulated world is so important, right?
>> Yes. Exactly. Because that would be the way to get one of the ways to get to it. What's the limits of what we can simulate? Because if you can simulate it, then in some sense, you've understood it.
>> I wanted to to finish with some personal reflections
>> of of what it's like to be at the forefront of this. I mean, does the emotional weight of this ever sort of weigh you down? Does it ever feel quite isolating?
>> Yes. Um, look, I I don't sleep very much, partly because it's too much work, but also I have trouble sleeping. It's very complex emotions to deal with because it's unbelievably exciting. Um, you know, I'm I'm basically doing everything I ever dreamed of. And we're at the absolute frontier of science on in so many ways, um, applied science as well as machine learning. And that's exhilarating as all scientists know that that feeling of being at the frontier and discovering something for the first time. And that's happening almost on a monthly basis for us. So, which is amazing. Um, but then of course we as as and Shane and I and others who've been doing this for a long time, we understand it better than anybody. Um, the enormity of what's coming and this thing about is still under actually appreciated. In fact, what's going to happen in more of a 10-year time scale. Um, including to things like the the phil philosophical uh, you know, what it means to be human, what's important about that? all of these questions are going to come up. Um, and so it's it's it's a big responsibility. Um, but we have an amazing team thinking about these things. Um, but also it's something I guess at least myself I've trained for my whole life. So, you know, ever since my early days playing chess and and then working on computers and games and simulations and neuroscience, it's all been for uh this kind of moment. Um, and it's roughly what I imagined it was going to be. So, that's partly how I cope with it is just training.
>> Are there parts of it that have hit you harder than you expected though?
>> Uh, yes, for sure. On the way, I mean, even the Alpha Go match, right? Just seeing you know that how we managed to to crack Go, but Go was this beautiful mystery and it changed it. And so, that was that was interesting and kind of bittersweet. I think even the the more recent things of like language and then imaging and you know what does it mean for creativity uh I I'm you know have huge respect and passion for the creative arts and having done game design myself and you know I talked to film directors and it's it's an interesting dual moment for them too. There's like first on one hand they've got these amazing tools that speed up prototyping ideas by 10x but on the other hand um is it replacing certain creative skills? So I think there's there's sort of these trade-offs going on um all over the place which um I think is inevitable with something as uh a technology as powerful and as transformative as as AI is as in the past electricity was and internet and we've you know we've seen that that is the story of humanity is we are tool making uh animals and that's what we love to do and for some reason we also have a brain that can can understand science and do science which is amazing. but also sort of insatiably curious. I think that's the heart of what it means to be human. And I think I've just had that bug from the beginning. And my expression of trying to answer that is is to build AI.
>> When you and the other AI leaders are in a room together, is there sort of sense of solidarity between you that that this is a group of people who all know the stakes, who all really understand their things, or or does the competition kind of keep you apart from one another?
>> Well, we all Yeah, we all know each other. I get on with pretty much all of them. Some of the others don't get on with each other. Uh and there is it's hard because that we're also in the most ferocious uh uh capitalist sort of competition there's ever been probably. You know, investor friends of mine and VC friends of mine who who were around in the dotcom era say this is like 10x more ferocious and intense than that was. In many ways, I love that. I mean, I I live for competition. It's it's it's you know I've always loved that since my chess days but stepping back uh I understand I hope everyone understands that there's a much bigger thing at stake than just you know company successes and and and you know that type of thing
>> when it comes to the next decade when you think about it are there big moments coming up that you're personally most apprehensive about
>> I think right now the systems are you know I I call them passive systems you you put the energy in as the user you know the question or the what's the task and then they uh these systems kind of provide you with some summary or some answer. Um so very much it's it's human directed and human energy going in uh and human ideas going in. The next stage is agent-based systems, which I think we're going to start seeing. We're seeing now, but they're pretty primitive. Like in the next couple of years, I think we'll start seeing some really impressive reliable ones. And um I think those will be incredibly useful and capable if you think about them as an assistant or something like that, but also they'll be more autonomous. So I think the risks go up as well uh with those types of systems. So I'm I'm quite worried about uh what those sorts of systems will be able to do maybe in two, three years time, you know. So, we're working on cyber defense in preparation for uh a world like that where maybe there's millions of agents, you know, roaming around on the internet.
>> And what about what you're most looking forward to? I mean, is there is there a day when you'll be able to retire sort of knowing that your work is done or or is there more than a lifetime's worth of work left to do?
>> Yeah, I always Well, I I could definitely do with sabbatical um and I would spend it doing stuff. Yeah, a week off for even even a day would be good. Um, but look, I think my mission has always been to get to kind of help uh the world steward AGI safely over the line for all of humanity. So, I think when we get to that point, of course, there's then there's super intelligence and there's post AGI and there's all the economic stuff we were discussing and societal stuff and maybe I can help in some way there. But I think um that will be my core part of my mission, my life mission uh will be done if it's a I mean it's only a small job, you know, just get that over the line or help the world get that over the line. You know, I think it's going to require collaboration like we talked earlier. Um, and I'm quite a collaborative person. So I hope I can I can help with that from the position that I have.
>> And then you get to have a holiday
>> and then I'll get I'll have the Yeah, exactly. a well- earned sbatical.
>> Yeah, absolutely. Deis, thank you so much. helpful as always.
>> Well, that is it for this season of Goo Deep Mind the podcast with me, Professor Hannah Fry. But be sure to subscribe so you will be among the first to hear about our return in 2026. And in the meantime, why not revisit our vast episode library because we have covered so much this year. From driverless cars to robotics, world models to drug discovery. Plenty to keep you occupied. See you soon.