Visions for the future with Lila Ibrahim | LSE Festival

Thank you, and thank you all for joining us today.

There are many decades in between. I will try to be concise. Yes, it has, you know, it's been quite a journey, and I think at many of the points along those decades, I'm not sure I would have told you I'd be living in London the past seven years as the first COO of an AI research company. But one thing that is consistent, a through line throughout, is this curiosity and drive to connect people and technology. So I studied electrical engineering and worked at Intel for 18 years, where I had, within a corporate environment during the rise of the personal computing industry, really had a chance to not just develop the microprocessor, the brain of the computer, but then eventually work with customers to design it in. And slowly I kept getting curious, and I found myself on actually how do computers and how does the internet fit into the world. And I remember a lot of being a part of that build-out actually shaped so much of my view of the work that I'm doing today, because seeing, you know, there was a time when we thought computers would take, what kind of jobs would they take, and did teachers still have a role, and what did this mean for the global economy. So it's really an amazing opportunity.

And then after 18 years, I had a chance to go into venture capital. And I thought, after 18 years, which is a very long time, I thought, you know, I can either stay here for the rest of my career or really go. You know, by that time I had worked in and lived in four countries. So I thought, now's the time to go and see what it is going from an entrepreneur, being very entrepreneurial in a corporate environment, to going and working with startups and entrepreneurs and sharing those best practices and also having a chance to learn. And I was blown away by the audacity of these entrepreneurs with these big ideas and what they could actually achieve. But what they didn't have is a lot of what I had been raised with, which is, you know, how is this going to impact the world? How do we think about doing this collaboratively? How do you build organizations? And I eventually realized that actually I'm a builder. And so I went into one of our portfolio companies that I helped make an investment in. Some of you may know Coursera. So I joined when it was just about 40 people and as the first president and COO, and really helped grow that business. And then I was, you know, several decades into the career and thought, I need a break. I just need, everything has been following a passion and interest. I had lived in Japan and Hong Kong and Shanghai. I had seen the rise of the personal computing, made investments in startups, built a startup. I'm like, I just need a break. And then I ended up here.

Oh, it's such a good question. I think throughout my life I've never quite felt like I fit in. You know, I was the foreigner in Indiana. Then when I lived in Japan as an exchange student in the '80s, I was the only foreigner in a city of 40,000. And when I was in electrical engineering, you know, being, I clearly looked very different than my classmates. And even English was my second language. So I think I always kind of grew up just saying, how do I, I needed to find ways to connect with people that were non-obvious and find ways to connect with different topics. But I think what that meant in many ways is that I saw things from a slightly different perspective. And while I was, during my time at Intel, what I found is I ended up in jobs where no one had done it before, and I didn't see that as a challenge, as a problem. I actually saw it as an opportunity because I was so programmed and used to that. And instead of trying to solve it on my own, I would definitely take a collaborative approach. And so I think that of like trying to seek other inputs and bringing it into the problem definition and into the solution really steered the journey that I had and, you know, really of how do you build things together and build it with people versus having the technology happen to people.

Yeah. When I was growing up, so my father was orphaned at the age of five or six, and he grew up in an orphanage where they really encouraged his educational development. So he, and when I was growing up, I saw my dad get out these really beautiful colored pencils and draw these beautiful pictures that actually turned into these tiny little computer chips that went into things like heart pacemakers. And so I grew up with this concept that electrical engineering solved problems and created solutions for, you know, added life. So how could this little orphan boy from Lebanon end up building technology that would save people's lives? And so to me, it was really the combination of art, math, and science. And those were topics I was very passionate about. And when I started at university, I had a chance to do some internships. And, you know, maybe if I had just done the theory, I like to go back actually to some of the women in engineering seminar classes because I tell them like, don't drop out. Like, what you're learning is the theory. What really matters is putting it into practice. And you're, you know, if you feel like you're not succeeding in some of your classes, that's okay. Like learning how to struggle and learning how to, like, what you do like and what you don't like is really important. And but what I got out of engineering too was this like asking questions of like why, to what end, and really trying to get to the heart of a problem. And I think that served me well.

Now, I have never had a formal business class other than through my Coursera work. I've never taken a formal business class. Everything I learned was on the job with people, asking questions, trying, succeeding, sometimes not succeeding as I had hoped, like failing and really having to learn from that. And I think that that grit, that resilience, and learning along the way. And it really reinforced that finding things that you're passionate about, because here's a slight secret that I can't believe I'm now saying online, but I actually thought I would go into electrical engineering without touching a computer. And so when I interviewed with Intel for the first time, I just remember my dad saying, "Don't tell them that you don't like computers." Which is really ironic when you think about my role now. So I do think that so much is going to change over a career and really embracing that change is so critical, and I think that's what my experience in early electrical engineering taught me.

Let's see. So in 2000, yeah, I was going to take a year off and reset and be really deliberate about what I wanted to do in this chapter of my life and my career. And my mentor from my venture capital days, John Doerr, said, "You've just got to meet this guy, Demis." And I'm like, "Who is he?" Like, why would I go to London to work on artificial intelligence? And we're here in the heart of Silicon Valley. There's plenty of opportunities if I want to do it. And besides that, I'm not a machine learning expert. And I finally decided as a favor to John, I would actually take the meeting. And I think the more I learned about AI, the more I realized this could be one of the most transformational technologies of our time. And there's a lot of things that could go wrong. And I think those 50 hours were them testing of like, okay, we're going to bring our first chief operating officer in. What does this mean in a research organization? And for me, it was a matter also of like, you know, you've got to really be, I want to be very deliberate. Is this what I really want to be doing? And in the end I decided that if I could bring all of my decades of experience bringing technology into market with communities in a responsible way, if I could help build some of that at DeepMind because I felt so values-aligned to the founders, Shane and Demis, I thought if I could do that then this would be a career chapter that was worth it. And so picked up the family and we moved here in the middle of, and my interviews were in January, which is also part of the 50 hours of like, do I really want to put myself through this? But it's been an amazing, amazing opportunity. And I, you know, my job is every day managing the risk while also making sure we're investing in the opportunity. And so those 50 hours really made sure that when I joined I knew what I was getting myself into, and every day that's what keeps me up at night. Are we doing enough to get the good stuff? Are we doing enough to mitigate the risks? And it has fulfilled my every hope and dream.

You know, when I... Easy question. Yeah. When I first started, like I said, it was really, I wasn't sure where and how I would be able to contribute to the mission of building AI responsibly to benefit humanity. But I knew it was something I really cared about in the values alignment with Demis and Shane. And the first thing I realized was, you know, you can't do a technical paper and then at the very end say, you know, you can have all the intentions at the beginning, but you can't get close to a paper review and say, are we going to release this? Is it, you know, have we thought enough about the downstream implications? So it was really about shifting to saying those values and ideals that we have, how do we actually build that in from the start of the research.

Now, what's interesting about DeepMind is, from my perspective, having been involved with a lot of technology companies, is this from the very beginning in London of pulling in a very interdisciplinary team where within a typical academic environment you might not have had people with those expertise interacting with each other. The social scientists with the technologists, the ethicists, the program managers. But within DeepMind, it was very much of like how the company was intentionally built. And my job was saying, okay, how do we actually leverage that as a strength and build that? How do we build these into the start of like even when we're starting to define what a research idea is, can we do it in partnership with the researchers? You know, they were looking at it very from a technical perspective or their area of expertise. But if we pulled together this interdisciplinary group and had the open conversations, and my job as a leader is to create the space to do that, to pull the people around the table, to create the safe space for those conversations and to allow the psychological safety to explore what could go right, what could go wrong, what skills, what voices are we missing and how do we augment that. So, you know, that was a very deliberate... I often talk about this, this is part of my job as COO that people may say like, that's not the sexy part about it, but it's actually one of the most critical things as a leader you can do. And, you know, when something doesn't go right, you have to be willing to take that accountability as well to create the space for those conversations to flourish.

You know, it's been seven years and we have really robust systems and we've done things like really invest in the technical safety, which was a pretty new field, to try to build this capability. But we also built up our group of ethics researchers and we strengthened our work around frontier safety, helped shape things like the Frontier Model Forum, a lot of our thinking around what are the dangerous capabilities, but also investing in the upsides as well of like, okay, great, we released a model like AlphaFold, but now it needs to get into the hands of researchers worldwide. How are we going to do that? And it's just not about making the paper or the model. So I would say that a lot of this comes down to, do you have the right responsible research in place, the right governance structures and the right path to impact. So those are all, and you know, what we're still learning, this is so important, is you need to have like, the Japanese word is kaizen, right? You think about, are we continuously improving? Are we making incremental? So it's not just releasing the models, but like learning when it's out, not only through our trusted tester community, but also more broadly in society and being able to iterate as we learn and as the technology continues to develop.

Yeah. And I think a lot of that, you know, just as we were talking about, of, you know, how you invest your resources. So that's, you know, your people, the talent that you want to hire. It's the budget to be able to do things, the management air cover. So I think when you look through our organizational structure of like what we have invested in, you'll see a continued investment in this space. And it's not just within our own organization. We've done a lot just realizing that, for example, we didn't necessarily have all the right voices around our table. Within, you can do the best you want to do within building an organization, but being able to do things like funding diversity scholarships to ensure that in the space of machine learning that we're bringing more voices and more fields, you know, the intersection of different fields, agriculture and NLP, or more in the social science as well, and just global coverage. So I think we've done a lot too to try to say, how do we help catalyze a lot of the work that's happening in this space, publishing even some of our own frameworks to share best practices with others in the community. We also have an opportunity to learn from others. But I do think this is an area we're continuing to, as AI capabilities grow, we're also continuing to invest in this space.

Yeah. And I have a positive bias on this because even from my Intel days, 15 years ago, I had a team of anthropologists and ethnographers as we were doing some of our work in developing countries. And I actually put them in charge of the architecture definition because they were thinking about how it was going to go out into the field. And so I think from that experience very early on in my career, which, you know, candidly, I didn't even know that was a field when I was in university. So it was like, I think one thing to remember as a social scientist is everyone's not necessarily aware of what skills and potential you can bring into the conversations. When I think about, you mentioned the ethics of advanced AI assistance, which was written in partnership with a group of social scientists. Like, I think this is such an important, the ability, what this skill can bring into the conversation. So my wish would be, help us remember the parts of what makes us human. What are the questions that we need to be asking even when it may not be comfortable to do so? And what are, because I think a lot of times we can easily get into, you know, from my experience in the tech sector, of like, it's easy to think you have a vision of how that technology might impact, but you need those different voices coming into the conversation to help shape it.

So I think things that we've been doing, for example, that I wish we could scale up, and I know this is not about any one company or any one country, like it really takes a collaborative approach, are things like when we're developing the technology, we know we're now coming at it from a technical perspective, but how do we bring the musicians along when we're doing generative music? How do we bring the movie producers along when we're doing generative video? The artists to think about things, the philosophers to help us ask the questions and for them to also ask the questions in a way that we know we don't have the answers. So it actually encourages us to actually expand the community of practice and the people that engage with us. And again, that's not just for us within Google DeepMind, but I think also more broadly for the sector. I think there are more questions than we have answers for as AI gets more general, more capable, more prevalent. And so I see the role of social scientists only increasing.

Yeah. Let's see. AlphaFold has been an amazing journey for us, and I even think back to when I started, it wasn't...

Always, it seems like it was always going to have a big impact, but when I started, it wasn't clear. And it's amazing that just seven years later, there's a Nobel Prize — the first Nobel Prize for the application of AI, which is significant. So AlphaFold is an advanced AI system developed by DeepMind that helps predict the 3D structure of proteins. So what that really means is that proteins are the basic building blocks of life. And if you think about diseases like Parkinson's or Alzheimer's, malaria — these are all protein-based diseases. Maybe you even remember, like, COVID and the spike proteins. So understanding proteins can really help us do things like understand disease, come up with better therapeutics. And how this used to work is you needed really fancy equipment, a lot of money, the right skill sets, and let's say a PhD's four to five years to understand how a single protein might fold.

What our advanced AI system did, AlphaFold, was do the prediction of all 250 million known proteins known to humankind, and is now available in a database. And this was basically — people thought it would take 50 plus years. We did it, like I said, within the past seven years, and we released it in a database. So it's as easy as, like, a Google Maps search, you can think of it. What that means is now researchers can more quickly get to understanding the protein and then do the research based on that. So it's — and they're using it for amazing things. Over 2.9 million researchers worldwide are using this, and candidly, we were surprised that there were that many researchers that were interested in proteins. And they're doing things like, why are some crops more resilient to diseases than others? University of Portsmouth is doing work around plastic-eating enzymes, so even dealing with industrial waste. And there's work with the Neglected Diseases Institute that we've done to ensure more equitable access of, like, okay, there's a lot of diseases that aren't necessarily being funded by pharmaceutical companies. So how do we ensure that we're also helping some progress in that space? So I think what's been really exciting is that we've now made a giant leap forward in using AlphaFold to advance scientific discovery. And this science symposium, as you mentioned, was to say, you know, the model is important, but what's really important is when it gets into the hands of the people who are experts and can go off and do amazing things with it. And so what can we learn from this that we might then be able to take into other fields, and how can AI become like a microscope or a telescope to help us understand the universe around us, such that we can, you know, address some of society's biggest challenges.