Google’s AI boss made OpenAI issue code red. Now he wants to solve disease. | Titans and Disruptors

0:00In 10, 15 years' time, we'll be in a kind of new golden era of discovery. That's why I hope a kind of new renaissance. We're in the thick of the AI revolution, but we might look back on January 2014 as one of the most pivotal moments in business history. That was the month that Demis Asabas sold his AI company, DeepMind, to Google. He rebuffed a higher offer from Meta's Mark Zuckerberg. And the acquisition scared Elon Musk so much that he decided to launch a rival company with Sam Altman, now called OpenAI. Fast forward to today, and Demis is still the one to beat. He runs all of Google's AI initiatives, including Gemini, which is quickly eating away at OpenAI's user base.

0:38In his spare time, Demis won a Nobel Prize, and he runs a startup called Isomorphic that wants to solve all disease with AI. I sat down with Demis at the World Economic Forum in Davos to learn where he thinks the future is heading. I'm Alison Chantel, and this is Fortune 500, Titans and Disruptors of Industry. We'll be back with Demis after a quick message from our sponsor.

1:17this is on the topic of every CXO conversation I'm a part of and I think the thought process has to be looking for high impact areas that may not be necessarily the most glamorous or high profile functional areas, but are ripe for automation and use of this technology to create efficiencies as well as innovation. And over time, AI agents will be also in customer-facing and growth-oriented domains. In our case, Deloitte, we're using it within our finance organization, looking at very mundane processes like expense management and working capital management. We're seeing other organizations using it in call centers and with software development that can be automated.

2:07Comes down to intentionality and so I think that intentionality in going functional area by functional area in concert with business and IT leadership in an enterprise, it needs to be a mainstream business planning effort that's budgeted, that's KPIs are developed, and there's real accountability for actual business outcomes and impact because of agentic capabilities. Demis, we're here at Davos. Thank you for making the time to do this with us. It's great to be here. You've had a huge 2025. It sounds like you're going up for a great 2026. But before we get into both of those things, I want to just take a step back so people can get to know you a little bit better.

2:47One of the things you love is chess. You're a chess master. You also love astronomy. And I'm curious how both of those things took you into AI or shape how you think about AI. Yeah. Well, I've always been interested in things like astronomy, cosmology, physics as a kid, because I've always been interested in the big questions. So, you know, what's actually happening here in the universe, consciousness, nature of consciousness, all of these types of things. So you get sort of drawn to physics if you're interested in the big questions. And then for me, for chess, I also love games, love strategy, ended up training my own mind by playing chess as a kid very seriously.

3:24And then that got me thinking about thinking and how does the brain work? And then I've combined all that together. That sort of led me to AI and computers and AI being a way to understand our own minds, but also a perfect tool for science and understand the universe out there. Equipped with a degree in computer science and a PhD in cognitive neuroscience, Hassabis co-founded DeepMind in 2010. The company launched with an ambitious goal, to solve intelligence. Under Hassabis' leadership, the DeepMind team made significant strides in its artificial intelligence models. And just four years later, Google purchased the research startup for hundreds of millions of dollars.

4:00You first started DeepMind, you co-founded it a number of years ago. and about 2014, you sold it to Google for about$500 million at the time. It was a hot deal. I know Meta wanted it too. And from my perspective, I think we're going to look back on that moment as one of the most transformative moments in business history. You've given Google the foundation of which to build an incredible AI machine and really take it into the future. When you look back on that, how do you feel about that moment? How did you make that decision? Did Did you know it was going to be such a big moment at the time? We did, actually.

4:36Those of us that were involved in the science. So it's interesting. We started DeepMind in 2010, which was 15 years plus ago now. And nobody was talking about AI. But we knew and we set out with a mission of solving intelligence and then using it to solve everything else. So we wanted to be the first company to build artificial general intelligence. And the main thing we wanted to apply it to was solving scientific problems. So when Google came along in 2014, and it was actually driven by Larry at the time, Larry Page, who was the CEO, we knew that in some ways we were sort of underselling. But on the other hand, what mattered to me was not the money.

5:10It was being able to, it was the mission and be able to accelerate our progress towards artificial general intelligence and answering these scientific questions that we were trying to solve. and I felt that teaming up with Google would accelerate that mostly because they had obviously enormous compute power and we see today that how important that is for developing intelligence. So at the time, I did mention to Larry and also the head of search at the time who was driving the deal that this would turn out to be, although it didn't look like it now, it might turn out to be the most important acquisition Google has ever done, which is saying something because they acquired YouTube and Android.

5:47They've got a good history of buying important things. And now if you go back and you look at the origins of open AI, if Elon and Sam got together because they were afraid that Google might now have a monopoly in the AI space with the DeepMind acquisition. So really that also kind of created a mega competitor at the time. Yeah, I guess there's all these sort of butterfly effects that happen. And I think partly also it was the success of things like AlphaGo, the first program to beat the world champion at the game of Go, using these kinds of learning systems that we're familiar with today, you know, reinforcement learning, deep learning at the heart of it.

6:21I think that was a big watershed moment as well in 2016. It's actually like the 10-year anniversary of that breakthrough this year. And I think that really started the starting gun for the modern AI era, including things like, you know, OpenAI. I know the founders of that watched that match and wanted a piece of that action. Following Google's DeepMind acquisition, the company repeatedly made headlines for its accomplishments in AI. In 2015, the company's AI model AlphaGo became the first computer to defeat a champion Go player. And later on, it defeated top players in chess, Stratego, and StarCraft II, which is a popular real-time strategy computer game.

6:56In 2020, Google's DeepMind AlphaFold II solved the protein folding problem. For decades, scientists had struggled to predict how a protein sequence forms in its final structure. The model accomplished this with remarkable accuracy. The team has since scaled its process to predict over 200 million structures, all of which are now available in an online database. His office received the Nobel Prize in Chemistry and a knighthood in 2024 in part for this achievement. Under Google and Alphabet, you've been able to have a lot of moonshots, take risks, try things that haven't necessarily led to money immediately, but have been profound breakthroughs.

7:31And one of them, you won a Nobel Prize. So I was wondering, congratulations, it's incredible. I was wondering if you could just tell me a little bit more about AlphaFold and why that's such a big deal in terms of how we could be looking at solving diseases moving ahead. Yeah, I think this is one of the benefits of being in as part of Google and Alphabet was having the resources and the time to really go after these sort of deep scientific problems. And Alphafold, I think, is the best example of that. It's basically a solution to a 50-year-old grand challenge in biology of can you determine the 3D structure of a protein just from its amino acid sequence, basically from its genetic sequence.

8:08and this is incredibly important because proteins basically do everything in your body from muscles to neurons firing everything depends on proteins and if you know the 3d structure of a protein what it looks like in your body then you kind of partially know what the function it does what it supports obviously it's important also for disease because things can go wrong with proteins they can fold in the wrong way like in something like alzheimer's and then that can create a disease so really important for drug discovery as well as fundamental biology and alpha folded was the solution to this problem that was posed 50 years ago by another Nobel Prize winner actually, Christian Anfinsen, that this should be possible and to go directly from a kind of one dimensional string of amino acid sequence to this 3D structure, this kind of how does it scrunch up into a ball.

8:53And AlphaFold was the solution and so efficient, not only is it accurate, we folded all 200 million proteins known to science and then we put that on a huge database with the European Bioinformatics Institute and for free into the world for everyone to use. So now over 3 million researchers around the world make use of AlphaVolt every day. Wow. And it's turned into, you're using some of it, I believe, for Isomorphic, which is a startup that you have, I don't want to say on the side. Yes. You're doing two huge jobs at once here. You've raised hundreds of millions of dollars and Google, of course, is a backer for Isomorphic.

9:25Can you just explain the mission there? And you have some lofty goals. Like you say, we're going to solve all disease. You don't say cure. You say solve. And also walk me through how hard it is to get a drug to trial, because that's historically been very difficult. So that was always the idea behind alpha fold. So obviously there's a lot of fundamental science that can be done if you understand the structures of proteins, including designing new proteins that do new things. So you can sort of use alpha fold in reverse to sort of go, okay, I want this particular shape. How do I get it from a genetic sequence?

9:56But to do drug discovery, knowing the structure of protein is only one small part of that whole process. And usually it takes like on average, you know, 10 years to go from understanding a target, you know, for a disease to all the way to a drug that's ready for in the market. So it's an enormous amount of time and cost, you know, billions of dollars, a decade or more. And most drugs, you know, fail along the way. It's only like a kind of 10 % success rate. So it's just incredibly inefficient because biology is so complicated. So what I've always dreamed about doing and was the first thing, you know, I wanted to apply AI to was human health, improving human health what could be more important use of ai and alpha fold was the proof point of that this could be possible and then isomorphic we spun that out after alpha fold was done so three four years ago to develop additional alpha fold level breakthroughs surrounding alpha fold so you can think more in the chemistry space so if you now know the structure of a protein you need to know where the chemical compound you're designing the drug basically where it's going to bind to the protein and what is it going to do?

10:59And so you need to build other AI systems that can predict all of that. So that's what we've been doing in isomorphic. It's going incredibly well. We have great partners with Eli Lilly and Novartis, the best farmers in the world. We have like 17 drug programs active already. And we plan to go to, you know, eventually be hundreds. And I think this is the way to make real step change progress in human health is you basically do your search and your hypothesis searching in silico and that's you know hundreds thousand times more efficient than doing it in a wet lab and you save the wet lab part just for the validation step of course eventually you have to test it in you know trials human trials and all those types of things to make sure everything's safe but you can do all of your search and design or almost all of it in silico that's the plan and so 2026 you mentioned is going to be a big year i imagine for both google and for isomorphic do you anticipate in early 2026 this could be the moment that you get the first drug to trial and made it be in cancer yes so we're working on actually several spaces of cancer cardiovascular immunology and then eventually we'd like to branch out to all therapeutic areas we're building a you know a general drug discovery engine platform you can think and uh we are already in pre-clinical trials very early stage for some cancer drugs and then you know hopefully by the end of the year if those who are successful will start going towards clinical trials.

12:24How do you manage yourself and your time and your teams? Because you're achieving really, really difficult things, whether it's the launch of Gemini 3, which was very successful and well-received, or it's getting drugs to trial. These are two sounding like very different things, two different teams to run. You can't be in two places at once. How are you doing this? How are you running two companies? One of my skills is bringing together amazing world-class interdisciplinary teams. I've loved managing those teams. I love composing those management teams together. And I've got incredible teams, both at Google DeepMind and at Isomorphic.

13:00And if we take Isomorphic, for example, we've blended top biologists and chemists along with top machine learning and engineering. And I think there's a lot of magic happens when you have these kind of interdisciplinary groups. And then if we think about on the Google DeepMind side, there we've tried to blend together the best of the startup worlds like what we were doing at DeepMind originally and then at scale you know in a kind of multinational scale with all the advantages of having these amazing product surfaces that we can immediately deploy you know technologies like Gemini 3 to and immediately get great feedback from users and also you know help in everyday lives of billions of users so it's amazingly exciting and motivating actually and in terms of the way I manage my time is you know I don't sleep very much but um like a couple hours yeah well if you know a bit more than that that would be bad for the brain so i do try and get six but i have uh unusual you know sleeping habits i sort of manage during the day and do try and pack my day in the office with as many meetings as possible back to back almost no time no break between then i get home spend a little bit of time the family have dinner and then i sort of start a second day of work about 10 p.m and go to 4 a.m where i do my thinking and kind of more creative work and research work.

14:14And it's worked out. You know, I've done that for about a decade now and it works well. I can't imagine being creative at four in the morning, but it's getting ready for you. Yeah, I come alive at about 1 a.m. In 2023, Google was facing increasing competition from other rapidly growing search engines and from the launch of ChatGPT. In the same year, Google pivoted to merge two of its AI teams, DeepMind and Brain, under Hassadis' leadership. And the goal was clear, Jumpstart the next generation of AI. You're clearly good at motivating teams to do hard things. I know in 2023, a decision was made at Google to put two different AI teams under you.

14:51How did you work out management kinks there and get the team shipping again? Because there was this feeling that Google was a little bit asleep at the wheel for AI. And I'm curious if you think that's true and how you got them to wake up. Yeah, well, we had two world-class groups in original DeepMind and Google Brain. And actually, I think often as a collector, we don't get enough credit for the fact that, you know, I think about 90 % of the modern AI industries built on technology or discoveries made by one of those two groups from Transformers to AlphaGo and Deep Reinforcer Learning. So we have, and we still have, I think, the deepest and broadest research bench.

15:28So we have an incredible talent, I think, better than anywhere else in the world by a long way. But it was getting complicated having two groups, especially given the amount of compute needed in this scaling era. So that was really why we had to put the two groups together so we could pull all of the talents together working on a single project in Gemini. But also even someone like Google didn't have enough compute to have two frontier projects under one house. So we needed to combine all of our reasons together. You know, I'm a very collaborative person. I'm very open minded about different ways of working and try to I'm always looking to improve as well.

16:05Like one of my watch words I live by is this Japanese word Kaizen that I love, which is sort of striving for continual self-improvement. And that's what I always try and do. I'm always in learning mode. Maybe perhaps that's why I like building learning machines because I like learning and there's always something you can learn no matter how expert you are at what you do. And bringing the two groups together and trying to combine the best of both cultures has been great. And I think we're reaping the rewards of that now. And now Google DeepMind is really, the way we think about it is like the engine room of Google.

16:34So we're sort of powering. It's like the nuclear power pump that's plugged into the rest of this amazing company in Google. And I think one of the things we did is one of the things I'm very proud of is getting the shipping culture going and sort of rediscovering, I guess, the golden era of Google back 10, 15 years ago and taking risk, calculated risk, shipping things fast and being innovative. And I think that's all working out really well now, whilst at the same time being thoughtful and scientific about and rigorous about what we put out in the world, whether that's engineering or scientifically.

17:05And I think and I hope, you know, we're getting that balance right. Yeah, and you mentioned kind of going back to the golden era of Google, so much so that the founders, at least Sergey, seems like he's back involved. How is it like working with him on AI and doing Google? It's been great. And Larry is too in different ways. Larry more strategically. Sergey's been in the weeds programming away, you know, on things like Gemini. And it's been fantastic seeing them engage. Are you putting him to work? Are you like, Sergey, I need this code right now. No, it's more like he chooses what to work on.

17:32But it's great seeing him in the office and pushing things in certain directions. and it's easier if the founders are kind of heavily involved. And I still act as well, like as a co-founder of Google DeepMind, of like as a kind of founder or two, right, in terms of like what we've got to do and strategically what we pick to do. And that's something I think I've learned to do well over the last, you know, 10, 15 years is when you have some ambitious goal, like solve all disease or build AGI, what are the intermediate goals that are also very ambitious, but that are kind of waypoints? What are the right ones to pick?

18:05And I think we've done that historically pretty well with most of the Alpha projects, AlphaGo, AlphaFold and so on, and then now Gemini. And I think that's really critical, actually, for any very ambitious scientific and engineering project is breaking it down into manageable steps so that you can see you're on the right direction. And I think that we very clearly are, I think, with the technology that we're building. And it's been an incredible couple of years for us. And I think we're getting into our groove, I would say. And I think other people and the external world are starting to feel that, you know, including things like Wall Street and the share price.

18:38In 2005, any questions about whether Google was facing an innovator's dilemma when it came to AI in search were answered. After the first quarter, its shares skyrocketed, driven in part by advancements it made in AI development, including the launch of Google's viral image generation model Nano Banana and Gemini 3. Alphabet's shares rose about 65 % by the end of the year, marking its best performance since 2009 and making it the top-performing stock among the Magnificent Seven. It definitely seems like there must have been some sort of KPI measurement, charge-ahead, unifying moment because, I mean, the launch of Gemini 3, much fanfare, among other launches, that caused OpenAI to go to a code red, which they claim happens all the time.

19:18Okay, sure. And then, you know, you have this huge monster deal with Apple that is monumental, I think, for the industry. So I'm curious, what happened internally behind the scenes? How did you set those KPIs for the team? And then how are you setting them to keep the momentum in 2026? Well, look, I think for me, it always starts with the research, like having the best models in this case, and obviously fundamental research feeding into that. And I always believe you then need to reflect that, obviously, as quickly as possible in your products, and then you've got to get your marketing distribution right.

19:48But none of it matters if your models aren't best in class and state of the art. And so that's what we focused on first with the Gemini models, but also our other models, things like Nano Banana, our image model, which went super viral. And that was a big part of their success last year. Our video model, VO, our world model. So there's more than just large language models. And we're kind of, you know, state of the art on all of those. And then it was about sorting things out internally, almost rebuilding the infrastructure in some way of Google so that you could reflect very quickly the power of the latest models into the Lighthouse products, including Search, YouTube and Chrome, all these amazing surfaces that we have, as well, of course, as the Gemini app.

20:28It was new for everyone in the industry. And I think, you know, it takes a little while to kind of re-architect things around that. And very much, you know, again, this idea of Google DeepMind being the engine room, providing the engine for the rest of the organization to use. And I think that took a year, 18 months to get right. But I think we're seeing the results of that now. And I think there's still more to go, by the way, and we can have even faster velocity. And I think the other thing is just also instilling this culture of intensity and pace and focus and really focusing only on the things that matter and kind of cutting out distractions.

21:05And then maybe the final thing I would say is I think there's a lot to say, especially in today's very noisy world, to just consistently deliver good decisions, good rational decisions. And over time, minimal drama. And then I think it's just amazing how much that compounds over time. And, you know, I think we're building a lot of momentum now. And I think hopefully we'll see that even more this year. Sort of like we mentioned before, the decision to sell DeepMind to Google was a monumental moment, transformational moment in business. If you're successful now, I think that will be perhaps the biggest transformation in business.

21:40How does that weigh on you to make sure that you as a leader are driving this in a direction that's good for society, good for the workforce, good for Google? because it is a little bit of an innovator's dilemma where this is the search king, huge business model based on ads. If you're successful, I don't know. Sure. Well, look, I mean, it is a classic innovator's dilemma. I think we've navigated it pretty well so far and search is more successful than ever. But also there's this aspect of like, if we don't disrupt ourselves, someone else will. So you're better off sort of being ahead of that, I think, and kind of doing it on your terms.

22:17and so i think that's what we found in terms of responsibility i feel that you know i've felt that way since not just at google but before that deep mind and before that even in my academic career because if we we you know myself and shane especially our chief scientists you know when we started deep mind we it seemed like a fanciful idea but we really believed that it would be possible to to create artificial general intelligence and we understood what i think more and more people understanding now is how transformative to the world that would be but also of course amazing for things like science and human health and maybe helping with energy and so on but also there are risks it's a dual purpose technology you know harmful actors you know bad actors could use it for harmful ends and eventually as agi becomes ai but technology becomes more autonomous more agentic and we get towards agi there's technological risk too and so i worry a lot about all of those things and i also of course you know we have to make sure that the the engine and the economic engine works as well so we have enough money to fund our research and fund things like alpha fold and give it to the world for free you know that's not easy it costs a lot of money to create something like and hire the researchers to create something like alpha fold but we do a lot of things like that and i want to do more things like that for the world but that requires us to be successful also on the commercial side so i think you know there's a balance to be had there but the responsibility i think part comes as well as and i feel like we can do this at Google as well is we have the platform to show how AI can be deployed in a responsible way and a beneficial way for all of society.

23:49And, you know, all of us who are Frontier Labs producing AI, we have choices about what should we use AI for? Are we going to use it for things like medicine and for alleviating, you know, administration and helping with things like poverty? Or are we going to use it for exploitative things? And I think that we're going to try and be a role model for all the good things that can come with AI. It doesn't mean we won't make any mistakes. We will do because it's such a nascent and complex technology, but we'll try and be as thoughtful as possible as we can with it. And we'll try and be as scientific about it as possible too.

24:22The scientific rigor we bring to our work and always have, I think is going to matter here a lot. I mean, it's a scientific endeavor in the end. And then, you know, I hope that what we, you know, the kind of reliability and security and safety that we like to work on will come through in our products. And then I think the market will reward that because if you think about enterprises that use these technologies as they get more sophisticated, they're going to want to know, you know, if you're a big bank or, you know, insurance company, whatever, health company, medical company, that you have some guarantees about what your AI systems that you're bringing in are going to do.

24:59And so I think that could be a good aspect of AI becoming very commercial is that there'll be sort of commercial incentives to be robust and reliable and secure and all the things that you'd want actually in preparation for HGI coming into the world. So when you look at the year ahead, what do you think the story of AI will be? What will we achieve? Well, I think this year, I mean, I say this every year, you know, every year is pretty pivotal in AI. And it feels like, at least for those of us inside, working at the coalface, you know, like 10 years almost happens every year. And I think this year will be no different it's very intense um but you've also got to kind of every now and again look up at the strategic picture i think that uh at least for us with gemini 3 we crossed the watershed moment in my opinion and and hopefully those of you who've used it will feel that in that it's very capable now and i i'm certainly using it in my everyday life to help me with my research and summarizing things and uh doing some coding so i think that the these systems are now ready to maybe build agents.

26:02We've talked, the whole industry has talked a lot about agents and more autonomous systems and delegating whole tasks to them. But I think maybe by the end of this year, we'll really start seeing that. I'm very excited about assistants coming into the real world with you, maybe on glasses. We have a big project on smart glasses. I think that the AI technology is only just about there to make that actually viable. And I think that could be a kind of killer app for glasses. You know, I think that part, bringing that into the world But also robotics, I think is going to, I still think there's more research to be done on robotics.

26:33But I think over the next 18 months or so, I think we're going to see kind of breakthrough moment in robotics, too. So all of these areas we're pushing very hard on, as well as, of course, improving Gemini itself. Those aren't the glasses, right? No, they're not. These are just normal ones for now. I would buy those if they were. I was going to ask you about the future form of how computers were not built for AI and all the things that AI can do. What do you think is the future for us? Sounds like glasses. I think Glasses would just be one of the solution. I have this side notion of, we talk about this notion internally of a universal assistant.