AlphaFold is an advanced AI system that predicts the 3D structure of a protein. We have 190 countries, over 3 million researchers using this to advance a number of fields and really unlock human potential while also being responsible.
The world's first and this world only chief AI...
Being brought into the AI research by Demis Hassabis himself as DeepMind's first COO. Today your title is called Chief AI Readiness Officer. In a world where this title probably did not exist maybe a year or two ago, what do you think readiness actually means for an individual and an organization?
I'm not sure the title exists beyond me.
I'm joined here today by a very interesting person who's actually seen technology evolve from inside out. You know, from microprocessors at Intel to expanding education through Coursera to being brought into the AI research, you know, by Demis Hassabis himself as DeepMind's first COO. Lila, thank you so much for coming here with us today. Your title at Google is called Chief AI Readiness Officer. I'd actually love to know what, according to you, does readiness mean both for an individual and for the organization.
Oh, thanks for having me here today. To me, readiness is really about how do we ensure that AI gets developed for and with communities and is not just something that happens to them. It was part of the reason I really wanted to shift out of being a chief operating officer. I had spent my career, and especially the past eight years, ensuring that we were building AI responsibly, that we were building the capacity within the organization to do that as well. And I feel like AI has gotten to the point now where the limitations are less technical and more about societal readiness. Are we developing the technology in a way that it can be useful? Are we mitigating the risks as much as we can? Are we ensuring that at a very local community that they're having the agency to apply this technology in a way that works for them? This all requires ensuring that humans are in the lead and humans are defining and shaping the future of AI. So readiness means about how do we ensure that there's access, that there's literacy, and that we're having the outcomes that we want to, and that we're learning all together so that we can continue on this journey of shaping a better future.
No, I think that's very interesting. I've been in this space where I'm trying to get companies ready in terms of AI, and the more we talk about AI readiness, I think that's where the space, the, you know, question comes from. I'd love to deep dive on this a little bit, but before that I want to take a step back. You know, we've seen you go from infra to education to being the COO at DeepMind, expanding DeepMind. What do you think is that thread that's connecting you all of these three together for you?
Yeah, I've been in technology for over 35 years, and I think in every step along the way I took on jobs and roles that were really interesting and curious for me, that I felt like I had something to learn and maybe something to contribute. Oftentimes I was like the first to do something, so it was very pioneering at the time. And the other through line is that it was always about how to make technology more useful for people, for everyday people in our everyday life. And although this every step I took, it never felt like I was making particularly conscious decision about it. When I look back and look at this portfolio of work that I have done, it really highlights to me that this circuitous route really leads me to the place I'm at with AI right now. If I can bring all of that expertise from these other industries and these other times in technology history into this moment in time in AI development to make sure that, again, that we're developing the technology from the beginning with the bigger impact in mind, then I think that's really important. So it's been, yeah, it all makes sense now in my mid-50s, but it didn't along the way.
Okay. So now if I were to look at what's happening today, right, when you talk about AI readiness and we're looking at organizations where you'd usually find some people who are probably experts, some people who are novices, and today you don't necessarily need to know how to code to build something. However, I think AI literacy is extremely important. How do you plan an AI rollout for organizations like these where you would have both novices and experts to a certain extent where it makes sense for everyone?
Yeah, it's a great question. I think one of the first things we have to think about is access. With a technology like AI, so many people have access even though they don't know they do. Then once they have access, we need to make sure they're literate, know how to use it, what it means to be a responsible user of this technology, what's their role, what's the role of the AI. And then just because you have literacy doesn't mean that you necessarily have the outcomes that you might want from using the technology. So I think it's not just about having access or literacy or like being able to do something with it. I think we're all in this continuous learning path because, you know, AI is advancing, continues to advance very quickly. The features advance, so we also have to keep up with what is it that we're bringing into this moment and what is it that the AI is able to do for and with us. So that's how I've been thinking about this. There's also, it's interesting if you kind of take a step back and say there's two parts of learning about AI. It's about how do we learn about AI as a topic? What is AI? How to trust it, when to use it, etc. And then there's AI as a tool. And so I think if we can think about how to partition these two of AI as a topic and AI as a tool, that that can help on someone's own readiness journey.
And then maybe one other thing I'll add is even within an organization like DeepMind where we're building AI, building AI responsibly to benefit humanity, but we also had to get ready ourselves. So there we looked at readiness as kind of a continuum. We had to ready ourselves within the organization of using the technology that we were building to learn from it and to have that user experience. We had to think about how do we ready governments, how do we partner with policymakers so they can understand this important work that they have to do around ensuring that it's safe and also creating opportunities for their citizens, and then how do we ready communities. So I think readiness can be looked at through many different lenses and it's not one size fits all. And it really should be a continuous conversation and continuous learning.
No, I think that's definitely true because we've been experiencing this firsthand with so many organizations. I think they start their journey with AI and they think once they start it's the end of it. But like you said, I think it's a very continuous process. But, you know, you spoke about DeepMind and I'd love to actually understand, this is a very personal curiosity that I have. What is it like working with technology for years that the public has not seen and then suddenly, you know, you go ahead, sort of release it for the public? You've open sourced so many things. Do you always expect them to act or react a certain way, but it doesn't probably work out that way? Do you happen to probably see them more excited, more scared, or just different than what you had imagined? Or at DeepMind, maybe this is not something you guys discuss at all? I'd love to know.
Well, we definitely discuss it, and I think whenever we're starting something like a new product or a new feature, we talk a lot about what the opportunity and the risks are, and we think AI could be such a transformational technology and it really requires exceptional care. The one thing that I will say I've learned over my time at DeepMind is that you cannot predict how humans will react. So a lot of what this means is we have to think about how do we bring in a diversity of voices throughout the development process, because by the time you're ready to deploy it, it's almost too late.
So, for example, when we're doing some of our technology, even in the science space, even though we might have around a table a diverse set of people, whether they're neuroscientists or ethicists or child psychologists or a computer scientist, we realize that we're still quite privileged. So we need to think about who are the organizations and communities we'll work with so that when it does get released, that we'll have already built through some of those learnings.
No, lovely. So because you spoke about DeepMind, and you know, DeepMind was able to solve something with AlphaFold that I think scientists had been looking for for years. Talk to me about what that day looked like. Was it one moment? Was it a couple of, you know, moments that came through where this entire breakthrough happened? And I'd love to sort of, you know, for you to go deeper into this point of when something like this happens, what changes in the organization, more from, you know, people perspective, and, you know, like because you mentioned risk, what happens there?
Oh, it's been an extraordinary journey. I've been at DeepMind nearly nine years, and I'll still never forget, I mean, there were times when we really struggled with our AlphaFold technology. And for those who might not know, that was the first time that the Nobel Prize was ever awarded for an application of AI, and it was to solve a 50-year grand challenge in biology. And really, it didn't start there. It actually started with something we called AlphaGo, which was, could we have a machine learn such that it could play the game of Go where there are more moves than there are atoms in the world. So never thought that this would happen for at least another five decades at least. So once that we made progress there, we were able to apply it to a real-world problem, which is AlphaFold. And like I said, it's not a straight line, research rarely is, and it was something we had to actually do with the scientific community.
Once we addressed what is called this protein folding problem, which is can you predict the 3D structure of a protein so that you could better understand diseases, come up with better therapeutics, you can deal with industrial waste, you can understand why some crops are more resistant to disease than others. So there's all these amazing applications of AlphaFold, and the journey we went on was, should we release it? Is it safe to release? How should we release it? So we now have it available in a database available to about 4 million researchers worldwide are using it. Nearly 200,000 are here in India using the database, and they're using it on areas that we didn't even think about, to be honest. So really surprised by both the number of scientists and the applications, and we've learned a lot in the process. Back to my point of like, we published the database in partnership with a European lab, and then we looked at the usage, who was using it, and we realized there were parts of the world that were not using it just because they had access. Again, it didn't mean they knew what to do with the technology, so we had to do a lot of programs in terms of trying to help scientists, even though they're scientists, weren't sure about using AI as a tool.
So it's been quite rewarding then to see the journey it's on. I would say for DeepMinders, that's what we call ourselves, it's been an extraordinary, humbling experience because this concept that AI could be like a microscope or telescope to help us understand the world around us and solve some of the world's biggest challenges, and it happened, and it meant that we learned so much because, like I said, we had to think about how to do it responsibly and with the ecosystem and the training and access, and we've been able to take that same type of model now that we've had the confidence to tackle other challenges like weather prediction.
And be able to work with the communities to be able to do things like predict hurricane paths 15 days in advance that can enable better crisis planning. So we've really been able to take that same type of model of like, what are these, we call them root node problems. What are these root node problems, challenges in the sciences that if we unlock it, that it really enables the entire scientific field to advance to provide better solutions for humanity?
No, I think that's also where I'd like to double down on the thing you said, which is you might have access to technology, but until you know how to use it and what to do with it, I think it's very difficult to predict how the humans would react to technology. To double down on, you know, AlphaGo and then Fold, I'd love to know, do you think things would be a little different today with existing AI tools and existing technology that we have? Do you think it'd be faster for DeepMind to maybe crack or invent something like this? Would it be a little different?
In those early days, there was something very special about not knowing what was possible.
And really pushing. And I think this is a multi-year challenge. I think we needed to tackle some of those really tough scientific challenges before we had even released some of the large language models because it helped us with the scientific method of really ensuring that we've tested something so rigorously that it's safe to put out into the world. I don't know, it's an interesting question to think about. I do think one of the things that hasn't changed though is that AI can just be a tool, and it's really the human ingenuity and creativity and human in the lead that helps to make some of these advancements, and clearly there are a lot more AI experts in the world today than there were eight years ago.
