[0:00:00] Welcome to Google DeepMind, the podcast with me, your host, Professor Hannah Fry. Now, when we first started thinking about making this podcast back in 2017...
[0:00:16] DeepMind was a relatively small, focused AI research lab. They'd just been bought by Google and they'd been given the freedom to do their own quirky research projects from the safe distance of London. Well, how things have changed. Because since the last season, Google has reconfigured its entire structure, putting AI and the team at DeepMind at the core of its strategy.
[0:00:41] Google DeepMind has continued its quest to endow AI with human-level intelligence, known as Artificial General Intelligence, or AGI. It's introduced a family of powerful new AI models called Gemini, as well as an AI agent called Project Astra that can process audio, video, image, and code. The lab is also making huge leaps forward in applying AI to a host of scientific domains, including a brand new third version.
[0:01:08] which can predict the structures of all of the molecules you'll find in the human body, not just proteins. And in 2021, they spun off a new company, Isomorphic Labs, to get down to the business of discovering new drugs to treat diseases. Google DeepMind is also working on powerful AI agents that can learn to perform tasks by themselves using reinforcement learning and continuing that legacy of AlphaGo's famous victory over a human in the game of Go.
[0:01:37] Now, of course, you'll all have been following this podcast since the beginning. You'll all be familiar with the stories behind all of those changes. But just in case you are coming to us fresh.
[0:01:48] Welcome. You can find our first award-winning previous seasons on Google DeepMind's YouTube channel or wherever you get your podcasts. They also, those episodes, go into detail about a lot of the themes that we're going to hear come up over and over again from the people here, like reinforcement learning, deep learning, large language models and so on. So have a listen. They are really good, even if we do say so ourselves.
[0:02:12] Now, all of the newfound attention on AI since the last series does mean that there are quite a few more podcasts out there for you to choose from. But on this podcast, in just the same way as we always have, we want to offer you something a little bit different. We want to take you right to the heart of where these ideas are coming from to introduce you to the people who are leading the design of our collective future. No hype.
[0:02:40] No spin, just compelling discussions and grand scientific ambition. So with all of that in mind, I am here with the DeepMind co-founder and now CEO of Google DeepMind, Demis Hassabis. Do I have to call you Sir Demis now? No, absolutely not. Okay, well, Demis, welcome to the podcast. Thank you. Thank you very much for being here. Okay, I want to know, is your job easier or harder now that there has been this explosion in public interest?
[0:03:09] I think it's double-edged, right? I think it's harder because there's just so much scrutiny focus and actually quite a lot of noise in the whole field. I actually preferred it when it was less people and maybe a little bit more focused on the science. But it's also good because it shows that the technology is ready to impact the real world in many different ways and impact people's everyday lives in positive ways. So I think it's exciting too.
[0:03:35] Have you been surprised by how quickly this has caught the public's imagination? I mean, I guess you would have expected that eventually people would have got on board. Yes, exactly.
[0:03:45] You know, those of us who've been working on it for, like us, for many years now, you know, even decades. So I guess at some point the general public would wake up to that fact and effectively everyone's starting to realize how important AI is going to be. But it's been quite surreal still to see that actually come to fruition and for that to happen. And I guess it is the advent of the chatbots and language models because everyone, of course, uses language. Everyone can understand language. So it's an easy way for the general public to understand.
[0:04:14] How would you describe these chatbots as though they were unreasonably effective, which I really like. And actually, later in the podcast, we are going to be discussing Transformers, which was the big breakthrough, I guess, the big advance that gave us those tools. But tell me first, what do you mean by unreasonably effective?
[0:04:35] What I mean by it is I suppose if one were to wind back five, ten years ago and you were to say the way we're going to go about this is build these amazing architectures like transformer architectures, of course, and then scale from there and not necessarily crack specific things like concepts or abstractions. These are a lot of debates we would have five, ten years ago is you need a special way of doing abstractions. The brain certainly seems to do that. But yet somehow the systems, if you give them enough data,
[0:05:04] Yeah, I suppose it is a surprise that...
[0:05:24] That things like conceptual understanding and abstraction have emerged rather than been embedded. Yes, and we would have been probably, we discussed last time, things like concepts and grounding. You know, grounding language in real world experience, maybe in simulations or as robots embodied intelligence, would have been necessary to really understand the world around us. And of course, these systems are not there yet. They make lots of mistakes. They don't really have a model of the world, a proper model of the world.
[0:05:53] a lot further than one might expect just by learning from language.
[0:05:58] I guess we probably should actually say what grounding is for those who haven't listened to Series 1 and Series 2. Because this was a big thing. I mean, we were talking about this a lot. Of course. So do you want to just give us an overview of what grounding is? So grounding is when, you know, there was one of the reasons the systems that were built in the 80s and 90s, the classical AI systems built in places like MIT, they were big logic systems. So you can imagine them as huge databases of words connected to other words. And the problem was you could say something, a dog has legs, right? And that would be in the database.
[0:06:27] But the problem was as soon as you showed a picture of a dog, it had no idea that collection of pixels was referring to that symbol. And that's the grounding problem. So you have this symbolic representation, this abstract representation, but what does it really mean in the real world, in the messy real world? And then, of course, they try to fix that, but you never get that quite right. And instead of that, of course, today's systems, they're directly learning from the data. So in a way, they're forming that connection from the beginning.
[0:06:57] In theory, there should be missing a lot of the grounding that you need, but it turns out that a lot of it is inferable somehow. Why in theory?
[0:07:05] Well, because where is that grounding coming from? At least the first kind of large language models don't exist in the real world. They're not connected to simulators. They're not connected to robots. They don't have any access to even, they weren't multimodal to begin with either. They don't have access to the visuals or anything else. It's just purely they live in language space. So they're living in a, they're learning in an abstract domain. So it's pretty surprising they can then infer some things about the real world from that.
[0:07:32] Which makes sense that the grounding gets in by people interacting with this system and saying, that's a rubbish answer, that's a good answer. Yes, so for sure, part of that is if the question that they're getting wrong, the early versions of this, was due to grounding missing.
[0:07:46] Actually, the real world dogs bark in this way or whatever it is, and it's answering it incorrectly, then that feedback will correct it. And part of that feedback is from our own grounded knowledge. So some grounding is seeping in like that, for sure. I remember seeing a really nice example about crossing the English Channel versus walking across the English Channel. Exactly, those kinds of things. And if it answered wrong, you would tell it it's wrong, and then it would have to sort of slightly figure out that you can't walk across the channel.
[0:08:13] Okay, so some of these properties that have emerged that weren't necessarily expected to be. I kind of want to ask you a little bit about hype. Do you think that where we are right now, how things are at this moment, is overhyped or underhyped? Or is it just hyped perhaps in the wrong direction? Yeah, I think it's more the latter. So I would say that...
[0:08:33] In the near term, it's hyped too much. So I think people are claiming can do all sorts of things it can't. There's all sorts of, you know, startups and VC money chasing crazy ideas that don't, you know, they're just not ready.
[0:08:45] On the other hand, I think it's still underhyped. Coming from you, Demis. Yes, I know, I know, I know. AI in 2010. Exactly, exactly. But, you know, I think it's still underhyped or perhaps underappreciated still, even now, what's going to happen when we get to AGI and post-AGI. I still don't feel like that's...
[0:09:04] People have quite understood how enormous that's going to be and therefore the sort of responsibility of that. So it's sort of both really. I think it's a little bit overhyped in the near term at the moment. We're kind of going through that cycle. I guess though, okay, so in terms of all of these potential startups and VC funding and so on, you who have lived and breathed this stuff for, as you say, decades, are very well placed to spot which ones are realistic goals and which ones aren't. But for other people, how can they...
[0:09:31] Distinguish between what's real and what isn't. Yeah, well, look, I think you need to look at, obviously, you've got to do your technical due diligence.
[0:09:40] Have some understanding of the technology and the latest sort of trends. I think also look at perhaps the, you know, the background of the people saying it, how technical they are. Have they just arrived in AI like last year from somewhere else? I don't know, they were doing crypto last year. You know, these might be some clues that perhaps, you know, they're jumping on a bandwagon. And it doesn't mean to say, of course, they could still have some good ideas and many will do, but it's a bit more lottery ticket like, shall we say. And I think that always happens when there's a ton of attention.
[0:10:10] And that creates a kind of opportunistic, shall we say, environment, which is a little bit opposite to the people, those of us who've been in for decades in a kind of deep technology, deep science way, which is ideally the way I think we need to carry on going as we get closer to AGI.
[0:10:39] So from the beginning with Gemini, we wanted it to be multimodal from the start. So it could process not just language, but also audio, video, image, code, any modality really. And the reason we wanted to do that was, firstly, we think that's the way to get these systems to actually understand.
[0:11:01] And so that's important. And we also had this vision in the end of having a universal assistant. And we prototyped something called Astra, which I'm sure we'll talk about, which understands not just what you're typing, but actually the context you're in. And if you think about something like a personal assistant or digital assistant, it will be much more useful if the more context it understands.
[0:11:30] about what you're asking it for or the situation that you're in.
[0:11:34] So we always thought that would be a much more useful type of system. And so we built multimodality in from the start. So that was one thing, natively multimodal. And then at the time, that was the only model doing that. So now the other models are trying to catch up. And then the other big innovations we had on memory, so like long context. So actually holding in mind, you know, a million or two million now tokens, you can think of them as more or less like words in mind. So you can, you know, give it war and peace or even a whole, because it's multimodal, a whole video now.
[0:12:04] Okay, and Project Astra, that's the new universal AI agent, the one that can take in video and audio data. At Google I.O., I think you used the example of how Astra could help you remember where you left your glasses, for instance. I wonder, though, about the lineage of this stuff, because is this just a kind of fancy...
[0:12:31] Advanced version of those old Google glasses. Because Google have a long history of developing glass type devices, actually back to, I think, 2012 or something. So they're way ahead of the curve. But maybe it was just missing this kind of technology. So you could actually understand a smart agent, a smart assistant that could actually understand what it's seeing. And so we're very excited about that digital assistant to, you know, to...
[0:12:55] Go around with you and understand the world around you. So it seems like a really, you know, when you use it, it feels like a really natural use case. Okay. I want to rewind a tiny bit to sort of the start of Gemini because it came from two separate parts of the organization. Tell me that story.
[0:13:11] Yeah, so we actually last year, we combined our two research divisions at Alphabet. So obviously the old DeepMind and then Brain, Google Brain into one, we call it a super unit, bringing all the talent together that we, you know, amazing talent we have across the company, across the whole of Google into one unified unit. And what it meant was, is that we combined.
[0:13:34] All the best knowledge that we had from all the research we were doing, but especially on language models. So we had Trinchilla and Gopher and things like that. And they were building things like Palm and Lambda and early language models.
[0:13:46] And they had different strengths and weaknesses, and we pulled them all together into what became Gemini as the first lighthouse project that the combined group would output. And then the other important thing, of course, was bringing together all the compute as well so that we could do these really massive training runs and actually pull the compute resources together. So it's been great. In a lot of ways, I mean, the focus of Google Brain and DeepMind was slightly different. Yes. Is that fair to say? Yeah.
[0:14:11] So I think it was, I mean, we were obviously focused on both of us on the frontiers of AI and there was a lot of collaborations already on a kind of individual researcher level, but maybe not on a strategic level. Obviously now the combined group, Google DeepMind, I kind of describe it as with the engine room, but it's worked really well. I think there were a lot more similarities actually in the way we were working than there were differences. And we've continued to...
[0:14:36] And keep and double down our strengths on things like fundamental research. So, you know, where does the next transformer architecture come from? We want to invent that. Obviously, we, you know, Google Brain invented the previous one. We combine it with deep reinforcement learning that we pioneered.
[0:14:50] I still think more innovations are going to be needed. And I would back us to do that just as we've done in the past 10 years, you know, collectively, both brain and deep mind. So it's been exciting. I want to come back to that merge in a moment. But I think just sticking on Gemini for a second, how good is it? How does it compare to other models?
[0:15:09] Yeah, well, I think it's, you know, some of the benchmarks are not, the problem is that we need more, I think this is one thing the whole field needs is much better benchmarks and benchmarks and capabilities. Well, there are some well-known benchmarks, academic ones, but they're kind of getting saturated now and they don't really differentiate the nuances between the different top models. I would say there's sort of three models that are kind of.
[0:15:30] So it's Gemini for Mars, OpenAI is GPT, of course, and then Anthropic with their Claude models. And then obviously there's a bunch of other good models too that, you know, people like Meta and Mistral and others built. And they're differently good at different things. It depends what you want, you know, coding, perhaps that's Claude and reasoning, maybe that's GPT and then memory stuff, long context and multimodal understanding, that would be Gemini. Of course, we're continuing to, all of us are improving our models all the time. So, you know,
[0:15:59] Given where we started from, which Gemini as a project only existed for a year, obviously based on some of our other projects, I think our trajectory is very good. So when we talk next time, we should hopefully be right at the forefront. Because there is still a way to go. I mean, there are still some things that these models aren't very good at. Yes, for sure. And actually, that's the big debate right now. So this last set of things kind of emerged.
[0:16:24] Planning in what sense? Well, like kind of long-term planning.
[0:16:40] So they can't problem solve something long term. You give it an objective. They can't really do actions in the world for you. So they're very much like passive Q&A systems. You know, you put the energy in by asking the question and then they give you some kind of response. But they're not able to solve a problem for you. You can't say something like if you wanted it as a digital assistant, you might want to say something like, you know, book me that holiday in Italy and all the restaurants and the museums and whatever. And, you know, it knows what you like. But then it goes out and books.
[0:17:09] So it can't do any of that. But I think that's the next era. These sort of more agent-based systems, we would call them, or agentic systems that have agent-like behavior. But of course, that's what we're expert in. That's what we used to build with all our game agents, AlphaGo and all of the other things we've talked about in the past.
[0:17:27] So a lot of what we're doing is bringing to kind of marrying that work that we're sort of, I guess, famous for with the new large multimodal models. And I think that's going to be, you know, the next generation of systems. You can think of it as combining AlphaGo with Gemini. Yeah, because I guess AlphaGo was very, very good at planning. Yes, it was very good at planning. Of course, only in the domain, though, of games. And so we need to sort of generalize that into the general domain of everyday workloads and language.
[0:17:56] You mentioned a minute ago how Google DeepMind is now sort of the engine room of Google. I mean, that is quite a big shift since I was asking in the last couple of years ago. Is Google taking quite a big gamble on you? Yeah, well, I guess so. I mean, I think...
[0:18:09] Google have always understood the importance of AI. Sundar, when he took over as CEO, said that Google was an AI-first company. We discussed that very early on in his tenure, and he saw the potential in AI as the next big paradigm shift after mobile and internet, but bigger than those things.
[0:18:27] But then I think maybe in the last year or two, we've really started living what that means, not just from a research perspective, but also from products and other things. So it's very exciting, but I think it's the right bet for us to kind of coordinate all of our talents together and then push as hard as possible. And then how about the other way around? Because I guess from DeepMind having that very strong research and science focus.
[0:18:52] Does becoming the engine room for Google now mean that you have to care much more about commercial interests rather than the sort of purer stuff that you do? Yeah, well, we do definitely have to worry more about, and it's in our remit now, the commercial interests. But actually, there's sort of a couple of things to say about that. First of all, we're continuing on with our science work and AlphaFold, and you just saw AlphaFold 3 come out, and we're doubling down on our investments there. That's, I think, a unique thing that we do.
[0:19:19] Google DeepMind now. And even our competitors point at those things as sort of universal goods, if you like, that come out of AI. And that's going really well. And we spun out Isomorphic to do drug discovery. So it's very exciting. That's all going really well. And so we're going to continue to do that. And then what was all our work on climate and all of these things. But then we're quite a large team. So we can do more than one things at once. We're also building our large models, Gemini and etc. And then we have a product team that we're building out that is going to
[0:19:48] Bring all this amazing technology to all of the surfaces that Google has. So it's an incredible sort of privilege in a way to have that there to plug in all of our stuff. And we invent something, it immediately can become useful to a billion people. And so that's really motivating.
[0:20:06] And actually, the other thing is, is there's a lot more convergence now between the technology you would need to develop for a product to have AI in it and what you would do for pure AGI research purposes. So there's not really, you know, five years ago, you'd have had to build some special case AI for a product. Now you can branch off your main research. And of course, you still need to do some things that product specific, but maybe it's only 10% of the work.
[0:20:30] So there's actually not that tension anymore between what you would develop for an AI product and what you would develop for trying to build AGI. It's 90%, I would say, the same research program. And then finally, of course, if you do products and you get them out into the world, you learn a lot from that, and people using it, and you learn a lot about, oh, your internal metrics don't quite match what people are saying. So then you can update that, and that's really helpful for your research.
[0:20:56] Absolutely. Well, OK, I mean, we are going to talk a lot more in this podcast about those breakthroughs that have come from applying AI to science. But I want to ask you about that tension that there is between knowing when the right moment is to release something to the public, because internally at DeepMind.
[0:21:15] Those tools like large language models were being used for research rather than being seen as a potentially commercial thing. Yeah, that's right. So we, you know, as you know, we've always taken responsibility incredibly seriously here and safety right from the beginning, you know, way back when we started in 2010 and before that. And Google then adopted some of our...
[0:21:36] Basically, ethics charts are affected into their AI principles. So we've always been well aligned with the whole of Google and wanting to be responsible about deploying this as one of the leaders in this space.
[0:21:45] And so it's been interesting now starting to ship real products with Gen AI in them. You know, actually, there's a lot of learning that is going on and we're learning fast, which is good because we're at relatively low stakes here with the current technology, right? So it's not that powerful yet. But as it gets more powerful, we have to be more careful. And that's just learning about the product teams that, you know, in other groups, learning about how to test.
[0:22:09] It's different from a normal piece of technology because it doesn't always do the same thing. It's almost like testing an open world game. It's almost infinite what you can try and do with it. So it's sort of interesting to figure out how do you do the red teaming on it. So red teaming in this case being where you're competing against yourselves almost. Yeah, so red teaming is when you set up a specific separate team from the team that's developed the technology to stress test it and try and break it in any way possible.
[0:22:37] You know, you actually need to use tools to automate that because nobody can red team. Even if you had thousands of people doing it, that's not enough compared to billions of users. When you put it out there, they're going to try all sorts of things. So it's kind of interesting to take that learning and then improve our processes so that.
[0:22:53] Our future launches will be as smooth as possible. And I think we've got to do it in stages where there's an experimental phase, then a kind of closed beta, and then launch. A little bit, again, like we used to launch our games back in the day. And learn at each step of the way. And then the other thing we've got to do, and I think we need to do more on, is use AI itself.
[0:23:14] to help us internally with red teaming and actually spotting some errors automatically or triaging that so that then the human, you know, our kind of developers and human testers can actually focus on those hard cases. You said something really interesting there about how you're just in a much more probabilistic space here, right?
[0:23:36] There's even a very small chance of something happening. If you have enough tries, eventually something will go wrong. And I guess there have been a couple of mistakes that, you know, public mistakes. Yeah, so that's why, you know, I think that, as I mentioned, that product teams are just getting used to the sorts of testing. You know, they tested these things, but they have this stochastic nature, probabilistic nature. So, in fact, a lot of cases where, you know, if it was a normal piece of software, you could say, I've tested 99.999% of things. So then it extrapolates. Yeah, so then it's enough because it's like, you know,
[0:24:05] There's no way of exposing the flaw that it has if it has one.
[0:24:09] But that's not the case with these generative systems. You know, they can do all sorts of things that are a little bit left field or out of the box, out of distribution in a way from what you've seen before. If someone clever or adversarial decides to, it's almost like a hacker decides to test push it in some way. And it could even be, I mean, it's so combinatorial, it could even be with all the things that you've happened to have said before to it. And then it's in some kind of peculiar state, which then, or it's got its memories filled up with a particular thing. And then that's why it outputs something.
[0:24:38] So there's a lot of complexity there, but it's not infinite. So there's ways to deal with it, but it's just a lot more nuanced than launching normal technology.
[0:24:48] I remember you saying, I think it was like in the first time I interviewed you, you said something really interesting about how actually you have to think that this is a completely different intelligence, a different way of computing. You kind of have to move away from the sort of the things that we completely understand, the deterministic stuff, into this much more messy, like probabilistic, error-ridden place, as well as your testers.
[0:25:13] Do you think the public slightly has to shift its mindset on the type of computing that we're doing? Yeah, I think so. Because, and maybe that's something we, you know, another thing, interestingly, that, you know, we're thinking about is actually putting out a kind of principles document or something before you release something to show what is the expectation from this system? You know, what's it designed for? What's it useful for? What can't it do? And I think, you know, there is some sort of education there needed of like,
[0:25:42] You'll be able to find it useful if you do these things with it, but don't try and use it for these other things because it won't work. And I think that that's something that we need to get better at clarifying as a field, and then probably users need to get more experienced on. And actually...
[0:25:58] It's interesting. This is probably why chatbots themselves came a little bit out of the blue, right? Even obviously ChatGPT, but even to OpenAI, it surprised them. And we had our own chatbots and Google had theirs. And one of the things was, is we were looking at them and we were looking at all the flaws they still had, right? And they still do. And it's like, well, it's getting these things wrong. And it sometimes does, you know, hallucinates and blah, blah, blah. And there's so many things. But then what we didn't realize is.
[0:26:22] Actually, there's still a lot of very good use cases for that. Even now that people find very valuable, you know, summarizing documents and really long things or writing awkward emails or mundane, you know, forms to be filled in. And there's all these use cases, which actually people don't mind if there's some small errors, they can fix them easily and saves a huge amount of time. And I guess that was the surprising thing. They sort of discovered, people discovered when you put it in the hands of everyone, there were actually these valuable use cases, even though this thing
[0:26:51] The systems were flawed in all of these ways we know. Well, okay. So I think that sort of takes me on to the next question I want to ask, which is about open source. Because when things are in the hands of people...
[0:27:01] As you mentioned, really extraordinary things can happen. And I know that DeepMind in the past has open sourced lots of its research projects, but it feels like that's slightly changing now as we go forward. Just tell me what your stance is on open source. Yeah, well, look, we're huge supporters of open source and open science. As you know, we've given away and published almost everything we've done, collectively including things like Transformers, right? And AlphaGo, we published all these things in Nature and Science. AlphaFold was open source, as we covered last.
[0:27:31] And these are all good choices. And you're absolutely right. That's the reason that all works is because that's the way technology and science advances as quickly as possible by sharing information. So almost always, that's a universal good to do it like that. And that's how science works.
[0:27:47] The only exception is when we have a dual-purpose technology, right? And AGI and AI, powerful AI, does fall into this. And so then the problem is that you want to enable all the good use cases and all the genuine scientists who are acting in good faith and so on, technologists, to build on the ideas, critique the ideas, and so on. That's the way society advances the quickest. But the problem is, how do you restrict access?
[0:28:13] at the same time for bad actors who would take the same systems, repurpose them for bad ends, misuse them, you know, weapon systems, who knows what. And, you know, those general purpose systems can be repurposed like that. And it's okay today.
[0:28:27] Because I don't think the systems are that powerful. But in two, three, four years time, especially when you start getting agent-like systems or agentic behaviors, then I think if something's misused by someone or perhaps even a rogue nation state, there could be serious harm. So then...
[0:28:45] I think that as I don't have a solution to that, but as a community, we need to think about what does that mean for open source? Perhaps the frontier models need to have more checks on them. And then only after they've been out for a year or two years, then they can get open sourced. That's sort of the model we're following with.
[0:29:00] Because we have our own open models of Gemini called Gemma. They're smaller, so they're not frontier models. So their capabilities are very useful still to the developer because they're also easy to run on a laptop because they're small in numbers of parameters. But the capabilities they have are well understood at this point, right? Because they're not frontier models. So it's just not as powerful as the latest, say, Gemini 1.5 models. So I think that's...
[0:29:24] Probably the approach that we'll end up taking is we'll have open models, but they'll be lagging, you know, maybe one year behind the most cutting edge models just so that those models that we can really assess out in the open, you know, by users, what those models can do, the frontier ones can do. And you can really, I guess, test those boundaries of the stochastic. Yeah, and we'll see what those are. The problem with open sources, if something goes wrong, you can't recall it.
[0:29:49] With a proprietary model, if your bad actor starts using it in a bad way, you can just sort of close the tap off. In the limit, you could switch it off. But once you open source something, there's no pulling it back. So it's a one-way door. So you should be very, very sure when you do that. Is it definitely possible to contain an AGI, though?
[0:30:09] Within the sort of walls of an organization? Well, that's a whole separate question. I don't think we know how to do that right now. So that's when you start talking about AGI level powerful, like human level AI. What about intermediary? Well, intermediary, I think we have good ideas of how to do that. So one would be things like secure sandboxing. So you test, that's what I'd want to test the agent behaviors in, is in a game environment or...
[0:30:33] A version of the internet that's not quite fully connected, right? So there's a lot of security work that's done and known in this space and in fintech and other places. So we'd probably borrow those ideas.
[0:30:44] And then build those kinds of systems. And that's how we would test the early prototype systems. But we also know that's not going to be good enough to contain an AGI, something that's potentially smarter than us. So I think we've got to understand those systems better so that we can design the protocols for an AGI. When that time comes, we'll have better ideas for how to contain that, potentially also using AI systems and tools to monitor the next versions of the AI system.
[0:31:11] So on the subject of safety, because I know that you were a very big part of the AI safety summit at Bletchley Park in 2023, which is, of course, hosted by the UK government. And from the outside, I think a lot of people just say the word regulation as though it's just going to come in and fix everything. But what is your view?
[0:31:30] Well, I think it's great that governments are getting up to speed on it and involved. I think that's one of the good things about the recent explosion of interest is that, of course, governments are paying attention.
[0:31:42] And I think it's been great. UK government specifically, who I've talked to a lot, and US as well, they've got very smart people in the civil service staff that understand the technology now to a good degree. And it's been great to see the AI safety institutes being set up in the UK and US, and I think many other countries are going to follow. So I think these are all good precedents and protocols to settle into, again, before the stakes get really high.
[0:32:07] So this is a sort of proving stage again as well. And I do think international cooperation is going to be needed ideally around things like regulation and guardrails and deployment norms. So because AI is a digital technology, very much so, it's hard to contain it within national boundaries, right? So if the UK or Europe does something, or even the US, but China doesn't, does that really help the world when we start getting closer to AGI? Not really. So I think my view in it is you've
[0:32:36] Because the technology is changing so fast, we've got to be very nimble and light-footed with regulation so that it's easy to adapt it to where the latest technology is going. If you'd regulated AI five years ago, you'd have regulated something completely different to what we see today, which is Gen AI. But it might be different again in five years. It might be these agent-based systems that are the ones that carry the highest risk.
[0:33:01] I recommend a sort of beef up existing regulations in domains that already have them, health, transport, so on. I think you can update them for AI, for an AI world, just like they were updated for mobile and internet. That's probably the first thing I do while doing a watching brief on making sure you understand and test the frontier systems. And then as things become clear and sort of...
[0:33:24] more clearly obvious, then start regulating around that. You know, maybe in a couple of years time would make sense. One of the things we're missing is, again, the benchmarks, the right tests for capabilities that what we'd all want to know, including the industry and the field is, at what point are capabilities posing some sort of big risk? And there's no answer to that at the moment, right? Beyond what I've just said, which is agent based capabilities is probably a next threshold. But there's no agreed upon test for that.
[0:33:53] You know, one thing you might imagine is like testing for deception, for example, as a capability. You really don't want that in the system because then you can't rely on anything else that it's reporting, right? So that would be my number one emergent capability that I think, you know, would be good to test for. But there's many, you know, ability to achieve certain goals, ability to replicate. And there's quite a lot of work going on on this now. And I think the safety institutes, which are basically sort of government agencies, I think it'd be great for them to do a lot, you know, to push on that as well, as well as the labs, of course.
[0:34:22] I wonder, in this picture of the world that you're describing, what's the place for institutions in this? I mean, if we get to the stage where we have AGI that's kind of supporting all scientific research, is there still a place for great institutions? Yeah, I think so. Well, there's sort of the stage up to AGI. I think that's got to be a cooperation between civil society, academia, government.
[0:34:48] And the industrial labs. So I think I really believe that's the only way we're going to get to the sort of final stages of this. Now, if you're asking after AGI happens, you know, that maybe that is what you're asking, then AGI, of course, one of the reasons I've always wanted to build it is then we can use it to start answering some of the biggest, most fundamental questions about the nature of reality and physics and all of these things and consciousness and so on. It depends.
[0:35:11] What form that takes, whether that will be a human expert combination with AI. I think that will be the case for a while in terms of discovering the next frontier. So like right now, these systems can't come up with their own conjectures or hypotheses. They can help you prove something. And I think we'll be able to prove, you know, get gold medals on International Maths Olympiad, things like that. But maybe even solve a famous conjecture. I think that's within reach now, but not.
[0:35:38] They don't have the ability to come up with Riemann hypothesis in the first place, right? Or general relativity. So that's really was always my test for maybe a true artificial general intelligence is it will be able to do that or invent Go, you know? And so we don't have any systems. We don't really know how even probably, you know, know how we would design in theory, even a system that could do that. You know, the computer scientist, Stuart Russell. So he told me that he was a bit worried that once we get to AGI.
[0:36:05] It might be that we all become like the royal princes of the past. You know, the ones who never had to ascend the throne or do any work, but just got to live this life of unbridled luxury and have no purpose.
[0:36:19] Yeah, so that is the interesting question. Maybe it's beyond AGI. It's more like artificial superintelligence or something. Sometimes people call it ASI. But then we should have radical abundance. And assuming we make sure we distribute that fairly and equitably, then we will be in this position where we'll have more freedom to choose what to do. And then meaning will be a big philosophical question. And I think we'll need philosophers.
[0:36:44] Perhaps theologians even to start thinking as social scientists that they should be thinking about that now. What brings meaning? I mean, I still think there's, of course.
[0:36:53] Self-actualization. And I don't think we'll all just be sitting there meditating. But maybe we'll all be playing computer games. I don't know. But is that a bad thing even or not, right? Who knows? I don't think the princes of the past came off particularly well. No, traveling the stars. But then there's also, you know, extreme sports people do. Why do they do them? I mean, you know, climb Everest. But I think it's gonna be very interesting. And that I don't know. But that's kind of what I was saying earlier about it's under.
[0:37:18] I think it will be very good in the limit. We'll cure lots of diseases and all diseases, you know, solve our energy problems, climate problems. But then the next question comes is, is there meaning? So bring us back like slightly closer.
[0:37:43] To AGR rather than superintelligence. I know that your big mission is to build artificial intelligence to benefit everybody. But how do you make sure that it does benefit everybody? How do you include all people's preferences rather than just the designers? Yeah, I think what's going to have to happen is, I mean, it's impossible to include all preferences in one system because by definition, people don't agree, right? We can see that, unfortunately, in the current state of the world. Countries don't agree.
[0:38:09] Governments don't agree. We can't even get agreement on obvious things like dealing with the climate situation.
[0:38:15] I think that's very hard. What I imagine that will happen is that, you know, we'll have a set of safe architectures, hopefully, that personalized AIs can be built on top of. And then everyone will have, you know, or different countries will have their own preferences about what they use it for, what they deploy it for, what they, you know, what can and can't be done with them. And that's fine. That's for everyone to individually decide or countries to decide themselves just like they do today. But as a society, we know that.
[0:38:44] There's some provably safe things about those architectures, right? And then you can let them proliferate and so on. So I think that we've kind of got to get through the eye of a needle in a way where as we get closer to AGI, we've probably got to cooperate more, ideally, internationally, and then make sure we build AGIs in a safe architecture way. Because I'm sure there are unsafe ways and I'm sure there are safe ways of building AGI. And then once we get through that, then we can sort of...
[0:39:11] Open the funnel again and everyone can have their own personalized pocket AGIs.
[0:39:16] If they want. What a version of the future. Okay, but then in terms of the safe way to build it, I mean, are we talking about undesirable behaviors here that might emerge? Yes, undesirable emergent behaviors, capabilities. Deception. Deception is one example that you don't want. Value systems, you know, we've got to understand all of these things better. What kind of guardrails work, not circumventable. And there's two cases to worry about. There's bad uses by...
[0:39:42] by bad individuals or nations, so human misuse. And then there's the AI itself, right, as it gets closer to AGI going off the rails. And I think you need different solutions for those two problems. And so, yeah, that's what we're going to have to contend with as we get closer to building these technologies. And also just going back to your benefiting everyone point, of course, we're showing the way with things like AlphaFold and Isomorphic. I think we could...
[0:40:08] You know, cure most diseases within the next decade or two if AI drug design works. And then there could be personalized medicines where it minimizes the side effects on the individual because it's mapped to the person's individual illness and their individual metabolism and so on. So these are kind of amazing things, you know, clean energy, renewable energy sources, you know, fusion or...
[0:40:30] I think they're all within reach. And then that would sort out water access because you could do desalination everywhere. So I just feel like.
[0:40:39] This enormous good is going to come from these technologies, but we have to mitigate the risks too. And one way that you said that you would want to mitigate the risks was that there would be a moment where you would basically do the scientific version of Avengers Assemble. Yes, sure. Terence Tao, get him on the phone. Exactly. Bring him on down. Yeah, exactly. Is that still your plan? Yeah, well, I think so. I think if we can get the international cooperation, you know, I'd love there to be a kind of international CERN.
[0:41:07] Basically for AI, where you get the top research in the world. Let's focus on the final few years of this AGI project and get it really right and do it scientifically and carefully and thoughtfully at every step, the final sort of steps. I still think that would be the best way. How do you know when is the time to press the button? Well, that's the big question because you can't do it too early because you would never be able to get the buy-in to do that. A lot of people would disagree. Today, people disagree with the risks, right? You see very famous people.
[0:41:36] I think you've got to get the timing right. I would say like these agent-based systems are going to be an interesting next step. So I think every time there's a big paradigm shift or big emergent new property comes along, you've got to take stock and think about, you know, what does that do to your timelines?
[0:42:06] DeepMind had unveiled this computerized rat with an artificial brain that helps to change our understanding of how the brain controls movement. But in the first season of the podcast, I remember we talked a lot about how DeepMind takes direct inspiration from biological systems. Is that still the core of your approach? No, it's evolved now because I think we've got to a stage in the last...
[0:42:31] I would say two, three years, we've gone more into an engineering phase, large scale systems, you know, massive training architectures. So I would say that the influence of neuroscience on that is a little bit less. It may come back in. So anytime where you need.
[0:42:46] more invention, then you want to get as many sources as possible. And neuroscience would be one of those sources of ideas. But when it's more engineering heavy, then I think that takes a little bit more of a backseat. So maybe more applying AI to neuroscience now, like you saw with the virtual rat brain. And I think we'll see that as we get closer to AGI, using that to understand the brain. I think it'd be one of the coolest use cases for AGI and science. I guess this stuff kind of goes through phases of like the engineering challenge, the intervention challenge. So it's done its part. It's, you know, for now.
[0:43:15] And it's been great. And we still obviously keep a close track of it and take any other ideas too. All of the pictures of the future that you've painted are still anchored quite in reality. But I know that you've said that you really want AGI to be able to peer into the mysteries of the universe down at the Planck scale. Yes. Like kind of subatomic quantum worlds.
[0:43:39] Do you think that there are things that we have not even yet conceived of that might end up being possible? I'm talking wormholes here. Completely, yes. I love wormholes to be possible. I think there is a lot of probably misunderstanding, I would say, still things we don't understand about physics and the nature of reality. And, you know, obviously the quantum mechanics and unifying that with, you know, gravity and all of these things. And there's all these...
[0:44:03] Problems with the standard model. So I think there's this and string theory, you know, I mean, I just think that... There's giant gaping holes in physics. Yes, in physics all over the place. And if you're, you know, talk to my physics friends about this. And there's a lot of things that don't fit together. I don't really like the multiverse explanation. So...
[0:44:21] I think that it would be great to come up with new theories and then test those on massive apparatus, perhaps out in space at these tiny, you know, the reason I'm obsessed with Planck scale things, Planck time, Planck space, you know, is because that seems to be the resolution of reality, right? That in a way, that's the kind of smallest quanta you can break anything into. So that feels like the kind of level you want to experiment on if you had powerful.
[0:44:47] Apparatus perhaps designed or enabled by having AGI and radical abundance. You need both. So to be able to afford to build those types of experiments. The resolution of reality. What a phrase. So as in like the resolution that we're at at the moment, sort of human level is just an approximation of reality. Yes, that's right. And then we know there's the atomic level and below that's the Planck level, which as far as we know is the smallest resolution one can even talk about things. And so that to me would be the resolution one wants to experiment on to really understand what's going on here.
[0:45:17] I wonder whether you're also envisaging that there'll be things that are beyond the limits of human understanding that AGI will help us to uncover that actually we're just not really capable of understanding. And then I sort of wonder if things are unexplainable or un-understandable, are they still falsifiable?
[0:45:36] Yeah, well, look, I mean, these are great questions. I think there will be a potential for an AGI system to understand higher level abstractions than we can. So again, going back to neuroscience, we know that it's your prefrontal cortex that does that. And there's sort of up to about six or seven layers of indirection one could take. This person's thinking this, and I'm thinking this about that person thinking this and so on. And then we sort of lose track. But I think an AI system could have an arbitrarily sort of large prefrontal cortex effectively. So you could imagine
[0:46:06] Imagine higher levels of abstraction and patterns that it will be able to see about the universe that we can't.
[0:46:10] Really comprehend or hold in mind at once. And then I think that from in terms of explainability point of view, the way I think that is a little bit different to other philosophers who've thought about this, which is like, we'll be like closer to an ant and then the AGI, right, in terms of IQ. But I think that's the way to think of it. I think we are Turing complete. So we're sort of full general intelligences ourselves, albeit a bit slow because we run on slow machinery and we can't, you know, infinitely expand our own brains.
[0:46:40] You can, in theory, given enough time and memory, understand anything that's computable. And so I think it will be more like, you know, Garry Kasparov or Magnus Carlsen playing an amazing chess move. I couldn't have come up with it, but they can explain it to me why it's a good move. So I think that's what an AGI system will be able to do. You said that DeepMind was a 20-year project. Yeah.
[0:47:04] How far through are we? Are you on track? I think we're on track, yeah, crazily, because usually 20-year projects stay 20 years away. But, yeah, we're a good way in now. 20 years is 2030 for AGI. 2030, yeah. So I think the way I say it is I wouldn't be surprised if it comes in the next decade. So I think we're on track. That matches what you said last time. You haven't updated your priors. Exactly. Amazing. Demis, thank you so much. Absolute delight. Absolute delight as always. Fun to talk as always as well. Thank you.
[0:47:36] Okay, I think there are a few really important things that came out of that conversation, especially when you compare it to what Demis was saying last time we spoke to him in 2022. Because there have definitely been a few surprises in the last couple of years. The way that these models have demonstrated a genuine conceptual understanding is one. This real-world grounding that came in from language and human feedback alone. We did not think that that would be enough.
[0:48:04] And then how interesting and useful Imperfect AI has been to the everyday person. Demis himself there admitted that he had not seen that one coming. And that makes me wonder about the other challenges that we don't yet know how to solve, like long-term planning and agency and robust, unbreakable safeguards. How many of those, which we're going to cover in detail in this podcast, by the way, are we going to come back to in a couple of years?
[0:48:33] And realised that they were easier than we thought. And how many of them are going to be harder? And then as for the big predictions that he made, like cures for most diseases in 10 or 20 years, or AGI by the end of the decade, or how we're about to enter an era of abundance, that sounds a bit like Demis is being overly optimistic, doesn't it? But then again, he hasn't exactly been wrong so far.
[0:49:01] You've been listening to Google DeepMinds, the podcast with me, Professor Hannah Fry. If you have enjoyed this episode, hey, why not subscribe? We've got plenty more fascinating conversations with the people at the cutting edge of AI coming up on topics ranging from how AI is accelerating the pace of scientific discoveries to addressing some of the biggest risks of this technology. If you have any feedback or you want to suggest a future guest, then do leave us a comment on YouTube. Until next time.