[0:00:00] Demis, thank you very much for your time. AlphaFold 3, big, big upgrade, of course, to the AlphaFold program, which was so significant in terms of the protein folding question for biologists and for the medical world. AlphaFold 3, so it's able then to predict the structures of all sorts of different biomolecules, including DNA and RNA, and how they interact with each other. Talk to us about the potential significant implications, what you see as the most significant implications of this technology. Yeah, we're very excited about this new AlphaFold version.
[0:00:28] Of course, biology is, as we know, is a dynamic system. Really, all the emergent properties of biology and life are due to interactions between different molecules and different structures. So that's what AlphaFold3 is about, is a first step in that direction, understanding this dynamic picture and understanding what proteins interact with and how they interact with those other types of molecules. So not just other proteins, but also DNA and RNA, like you said. And for the medical world and in terms of drug discovery,
[0:00:56] How significant do you think this could be in the months and years ahead? Well, of course, the holy grail for drug discovery is not just knowing the protein structure, which is what AlphaFold2 did, but actually designing...
[0:01:06] drug compounds called ligands that bind to the protein surface and you want to know where it binds and how strongly it binds in order for you to design the right kind of drug compound. So AlphaFold3 is a big step in that direction of predicting protein ligand binding and how that interaction will work. Given the technology you now have at your fingertips and everything you know about how this is progressing, what is your best guess as to what will be the first AI
[0:01:35] Derived and discovered drug. Yeah. Well, we announced earlier on this year some big partnerships with Eli Lilly and Novartis. So we're already working on real drug programs. And I would be expecting maybe in the next couple of years, the first AI sort of design drugs in the clinic. That's really interesting. In the next couple of years. Isomorphic Labs, look, I know this is a passion of yours. You oversee this unit. It's a separate business, but under the Alphabet family, of course.
[0:02:04] And it's that generative AI or AI developed drugs. And you have those partnerships, as you say, Novartis, I should say, and Eli Lilly.
[0:02:13] Up to $3 billion of revenue. Does AlphaFold 3 get you closer to unlocking that revenue with those commercial partners? Yeah, it's a critical part of the tools that we have at Isomorphic. We also have some other AI systems that fill out this space of drug discovery. There's a lot of different techniques and modules that are needed in order to cover the whole of that very difficult process. AlphaFold 2 and the protein structure was just one small component of that. And you can think of Isomorphic building many tools that fill that out.
[0:02:42] Moving more into things like biochemistry and chemistry so that we can design the right drug compounds. One of the elements that stood out to me was RNA. And when we think of RNA, a lot of us will think of mRNA and the successful vaccines against COVID. What are the potential?
[0:02:56] What are the future adjustments or the implications for mRNA vaccines from this technology? Yeah, I think this is a really interesting space that we can now get into, I think, with AlphaFold3 and the new capabilities it has into sort of biologics, maybe antibodies, this sort of new space beyond small molecules. And we think there's great advances to be made there as well as with chemical compounds and small molecules. So I think it sort of opens up the type of opportunities that we can tackle.
[0:03:23] You are obviously a key figure when it comes to the generative AI and what's evolving. And within Alphabet now, all of the AI elements have now been put under your control. Gemini, DeepMind, of course, Isomorphic Labs, all the research, all the compute. Does that, do you think, better equip DeepMind, Google DeepMind and Alphabet to fight back, beat back the competitive threat maybe from the likes of OpenAI?
[0:03:48] I think it's part of the exciting era that we're entering into with what AI can do in products as well as science. And I think the amount of compute and engineering resources and other things that are needed, I think the new kind of organizational setup we have streamlines what we need to do and sets us up brilliantly for the next exciting phase. And positions you to compete better.
[0:04:12] Well, I think it will allow us to go faster and produce more and basically continue with all the amazing advances that we've been doing over the last decade. And you've talked about being in a low data regime. And when we think about what's needed for generative AI, huge amounts of data, of course, and for some, not all, but for some copyrighted data, but also huge amounts of energy, electricity, water, and huge amounts of cash.
[0:04:36] Are the foundations of generative AI challenged when it comes to the sustainability of this? I think that the benefits that the generative AI models that we build, things like drug discovery and other stuff that is going to far outweigh these costs, I think in the long run. So including even on things like energy and climate, where I think AI has a huge potential to help with that.
[0:04:59] Derive more efficiency from our power grids, inventing new materials, new technologies, clean technologies, things like helping with fusion. I think that AI can be applied to many, many of these areas and be extraordinarily productive and helpful in all these areas and far outweigh the costs and efforts that are involved in building these models. So it can be sustainable and you don't see these as major roadblocks. I mean, particularly the energy component and the shortage of data now.
[0:05:27] Well, look, there's challenges. There always is in research. And we've got to be innovative to figure out our ways around that and how to do that. For example, the data question, there's lots of interesting work going on with things like synthetic data and how to self-generate data. And we're actually experts in that from the past with our gaming work, where if you take our programs like AlphaGo, that was self-learning, self-improving through playing against itself. So in effect, it was generating its own data.
[0:05:56] And I think some of those kinds of techniques can be transferred over to the new world of generative models and large language models. Broadly across what's happening with generative AI, the critics, the skeptics say, look, there's a lot of hype involved. And we have seen companies like Stability AI, they've been challenged with their finances. The likes of Inflection being absorbed by Microsoft. Are we at a point now where we're seeing a little bit more rationalization, where we should look to maybe see more closures, more failures, more acquisitions?
[0:06:26] I think, funnily enough, in many ways, I think AI is overhyped in the short term and probably underestimated over the long term, like what it's going to bring. And I think that's probably true of a lot of breakthrough technologies. So I think we're seeing, because of the popularity of AI now in the last couple of years, lots of people trying to get into this space who maybe haven't thought about this as long as people like us who've been in it for decades. And I think we're going to see a sort of rationalization process happening.
[0:06:55] I actually think what it's going to end up delivering is going to be even beyond what the most optimistic end of things that we can see are in the near term. But in the longer term, I think there's going to be enormous benefits and enormous opportunities. But as with any new market and new technology, there'll be a lot of twists and turns on the way there. You were involved, of course, in the UK AI Safety Summit six months ago now. Is the UK still moving at pace when it comes to AI, or are we slipping behind?
[0:07:24] I think it is moving at pace. And I'm very pleased to see government and civil service sort of embracing that. And I think the summit was a huge success. And obviously, there's the next one in Korea and then the one next after that in Paris. But the UK specifically? Yeah. There's more that we can do. And there's examples of a data center that plans to build it on the outskirts of Oxford. And it was essentially nixed by the Secretary of State in charge because it essentially didn't look good.
[0:07:52] We need those data centers would be the proponents of generative AI and we need to reform our planning laws. Is that a roadblock for the UK? Yeah, I mean, I don't know about that specific case, but in general, we've got to make sure we embrace the enormous opportunity that AI represents.
[0:08:08] The enormous economic opportunity for startups and business and embrace that quickly and get on board with that quickly, as well as also for our infrastructure and services. Things like the health service, I think, can benefit from these technologies and allow them to be more efficient and more performative and more productive. So I think there's enormous opportunity. We have the intellectual capital here. We have lots of leading companies here. We need to make the most of that.
[0:08:38] And I think government needs to embrace that as well as the responsibility side, which I think the summit was about too. So I think both are important and we shouldn't lose sight of either. Big election year, of course, for the UK. I'm not going to push you on your political colours, Demis. But do you think, are you confident that the Labour Party, who of course have a huge polling advantage right now, are invested, as invested in AI as the Prime Minister Rishi Sunak claims to be?
[0:09:01] I feel like from the conversations I've had, both parties understand this enormous opportunity and the responsibility that goes along with that that's coming up and the incredible potential AI has. So I feel pretty confident about the discussions I've had with officials from both parties. As we look to the end of the year, so much changes within this space.
[0:09:23] So much evolution, so much innovation. What are the kind of changes, what are the big things that you think are going to come about between now and the end of this year that you've got on your radar? Well, look, I think there are huge things in the science space. I'm very excited about what isomorphic is doing and what AlphaFold3 can do with drug discovery.
[0:09:41] I'm also really excited about the next stage of these large general models. I think the next thing we're going to see perhaps this year, perhaps next year is more agent like behavior. So systems that are able not only to just answer questions for you, but actually plan and act in the world and solve goals. And I think those are the things that will make these systems sort of the next level of usefulness in terms of being a useful everyday assistant. So those AI agents are maybe
[0:10:11] One to two years away in terms of having that utility. Yeah, I mean, we're working hard and others are too. And again, that's bringing back in some of our work we did years ago with gaming, which were all agent systems, you know, achieving certain goals and objectives, bringing that kind of work and marrying that together with the modern large multimodal model work. Bringing it back to isomorphic labs, because this is where a lot of this started in terms of drug discovery and the conversation there. Are you under pressure when it comes to isomorphic labs to commercialize
[0:10:41] What is the potential revenue stream of that unit?
[0:10:44] Well, I think first and foremost, I think it's the most beneficial thing. If you ask me the number one thing AI can do for humanity, it will be to solve hundreds of terrible diseases. I can't imagine a better use case for AI. So that's partly the motivation behind Isomorphic and AlphaFold and all the work we do in science is to advance society and benefit society in these big ways. But I also think that if you could revolutionize drug discovery process.
[0:11:14] Make it 10x faster and more efficient and more likely to pass through the clinical trials because you can predict the properties better. That has to also be enormously of an enormous commercial value too. So I hope to do both with isomorphic. Build a multi hundred billion dollar business. I think it has that potential as well as be incredibly beneficial for society and humanity. And on the approvals process, clearly really essential when it comes to how drugs get to market.
[0:11:41] Does the AI generated drug story, does that lead to a faster approval process or does it actually extend?
[0:11:48] The approval process. I think to begin with, it will just go through the normal approval process, which is itself quite slow. But I mean, we're focused on the discovery piece, like getting to the clinic and that's in itself a very challenging, slow process. So we want to improve that. But I think at the next stage, you know, once people get used to the efficacy of these AI design drugs and perhaps a few of them have been proven out and go to market and then become super beneficial for patients, then we can, you know, at some point look at that whole process and see.
[0:12:17] Let's see if things can be more optimized there. Demis, before I let you go, what keeps you up at night? Well, I don't get much time for sleep with all the things going on. I'm sort of too excited. And also, you know, there's enormous responsibility as well with bringing these incredibly transformative technologies into the world. So, you know, it's a mixture of many things. There's always something going on in the world of AI. Okay. Demis, thank you very much indeed.