This week, I talked to Google DeepMind Chief AI Readiness Officer Lila Ibrahim. Amid concerns about the dangers of AI, she's working to make AI a force for good, to solve the hardest technological challenges of our time and to improve billions of people's lives. We talked about the importance of focusing on responsible development from day one.
There are risks. There are opportunities. We have to work together. So, how are we going to go about doing that? I think of myself very much as a builder, and I don't think I could do this job if I didn't believe that the benefits will outweigh the risks.
And she shared her optimism, profound insights, and practical advice. Take a listen.
There is so much happening in the AI space right now, and you are at the epicenter of it, at the cutting edge of all the technology that we are talking about all the time. Lila, can you start off by just explaining what is Google DeepMind and what is your role there?
Oh, and thank you so much for having me today. So I joined DeepMind about eight and a half years ago as the first chief operating officer, and really the mission is around how do we build AI responsibly to benefit humanity. And I'd say the majority of my time at DeepMind has really been on that first part of like, how do we build AI responsibly? What does responsibility mean? How do we build it into the ethos of what we do and how we do things? Then the past few years, I've really been thinking about AI has gotten good enough and solid enough that technology advancements that we can think about how do we really start applying this to real-world problems. That's the to benefit humanity part.
Earlier this year, I announced that I was shifting from my chief operating officer role to becoming the chief AI readiness officer. I think I might be the first one in the world, and my feeling is if I do this job right, then other companies will follow. And the whole concept is how can I shift my time from focusing on building AI to how do we actually responsibly work with communities to get the benefits of this technology, so that technology, AI, happens with communities and not to communities.
So the AI readiness is not Google DeepMind's AI readiness. Your job is to help the rest of the world be AI ready.
Yeah, exactly. I focused on the AI readiness of the organization in my past life, and this was really a chance for me to start taking all of those best practices and saying this technology is coming fast and it is powerful, and it's really important that we bring communities along and learn together to shape a better future. You can't shape the right future unless you do so with the communities that it's going to impact.
What is so incredible about your work at Google DeepMind, and you've been in this industry for 30 years, is that you were working at the cutting edge of AI before the ChatGPT moment. And ChatGPT is when we all started playing around with AI and it became part of the everyday lexicon, and generative AI really started to impact our lives. What is the purpose of Google DeepMind, especially within the parent company Google? I mean, we all use Google's products every day. What is Google doing with DeepMind, and why does it make sense?
It might help if I start by giving a little bit of context around DeepMind. DeepMind was founded in 2010. I'm based here in London, and it was founded just around the corner. It was really this concept that if we can use AI as a tool, such as a microscope or a telescope, to understand the world around us, then we can come up with better solutions to solve some of humanity's biggest challenges, whether it's addressing diseases or industrial waste and pollution, etc. In 2014, the company was acquired by Google, remained largely independent, but still based here in London. I joined in 2018, and shortly after, in 2023, is when we took parts of Google, we consolidated all of the AI activities, merged it with DeepMind to form Google DeepMind.
What DeepMind is today is the engine room of AI innovation for all of Google products. So you may know Gemini, which is developed here in our office here at Google DeepMind. But it's not just the large language models and this generative AI. We also do work on how do we continue to advance the field of AI. You know, the T is about transformer technology, which was started at Google. So a lot of the modern AI era actually has much of the innovation that both DeepMind and Google had developed. So we've continued to invest in how do we think about things like memory and context, such that we can continue making AI models more factual and useful for human beings. So we do that type of foundational research.
And then the other area that we really focus on is scientific problems. If we can apply AI to solve some of our biggest challenges, whether it's how do we think about weather forecasting, how do we think about protein folding to understand protein-based diseases to come up with better therapeutics, whether it's even applying AI to ancient texts to understand what happened millennia ago that we don't understand. So AI has, we think a lot about how do we understand what the AI-shaped problems are and then apply it to what we call a root node problem. So if we can open up an entire field for scientists, material scientists, biologists, to then be able to use that tool to advance their fields directly.
That's so interesting. I mean, obviously the Gemini example is something we could all relate to. We all search with Gemini every day. But this idea that you're trying to solve these root node problems that could unlock other research. Protein folds. I don't know what a protein fold is, but I understand that that would impact drug development. Or even the ancient text examples. So interesting. What does that mean in terms of your office? Because I heard you have quite an eclectic mix of experts working around you.
Yeah, in fact, from day one, we've always thought about how do we have an interdisciplinary team because that's really this messy intersection is where ideas can really flourish. And one of the ideas when we first started was, if you know, at the time a lot of the AI expertise were in universities, in academia, they get kind of siloed in their individual departments. By being in an organization, we can create the culture and the expectation of the interactions of them. So, how do you have the neuroscientist with the biologist and the ethicist and the policy lead sitting around a table talking about the potential impact of the next generation of technology and what are the responsible guardrails we're going to put in place.
So, before we talk about responsibility, because I think that's a very important topic, you talked about how your new role, shifting from COO to this new whole category of roles of AI readiness officer, which is now more outward-facing. You're talking about bringing solutions to the real world, but how do you at Google DeepMind figure out which node problems, which sort of central issues you want to tackle? Whose idea is it to get some ancient text and try to use AI to decipher it?
Yeah, it's really an interesting question, and maybe it varies by sector. And in many ways, it depends a lot on the talent that we have in the organization. We hire smart people who know the technology well and like to ask questions and be curious, and we know that we also don't have all the answers. So we have to go out and meet with different research organizations to understand what they're looking at. So take weather, for example, because I think globally that's a challenge. If we can better predict weather, then we can better mobilize emergency crews, for example. In many ways, you think weather is not that predictable, but if we can partner with weather agencies globally, then we can think about working together on the data and make this available to everyone.
We're now able to predict 15 days in advance 50 different permutations of how a hurricane might evolve. And in fact, we were able to accurately predict Hurricane Melissa. Now, when you think of what this might mean for, like, a, that could be huge. The ancient text was like a very personal story of a researcher who was sitting around a table with some friends, and, you know, they're like, how do we, where we have parts of the stone that is missing text in different languages, and we have to fill this in. Could generative AI be helpful? And it was really just more of an opportunistic.
AlphaFold, a 50-year problem in the field, like unsolvable. We were able to take what was essentially a billion years of research and make it available in a database of 250 million known proteins. And you remember, like in COVID, it was all about the spike proteins, right? So proteins, and you can predict how they fold or misfold, who can have a better understanding. That's expensive equipment. It's a lot of know-how. Not, researchers worldwide don't have access to this. So, by being able to provide it readily available, we're seeing really innovative uses, including things like dealing with diseases that are rare that maybe don't have enough funding to get the support for the type of research that now it's readily available.
And as you mentioned, AlphaFold is an open platform, and my colleagues were recognized with the Nobel Prize a couple of years ago for their work in the technical breakthrough in making that available.
So it seems like accessibility has always been central, accessibility of AI has been central to the mission of Google DeepMind, or accessibility of this technology. Now it's really your focus. How do you see the potential around accessibility, or sort of what you're trying to do now in this new role, changing the way DeepMind's technology is accessed and even the kinds of impact it might have globally?
Oh, I'm really happy that you narrowed in on this question. This is something that's actually really important to me as a human being and as a leader. Like, I often reflect on, we're all in our leadership roles in this moment in time in history, and we have this responsibility to shape a better future. And I think a lot about this as a parent, you know, with my teenage daughters. And for me, having spent earlier in my career with the computing industry, being at the center of that, when computers were available, we had to think about internet and wireless connectivity, or even when social media first came out, the lessons that I learned through those generations really helped influence my thinking of what I'm trying to do in this role.
Accessibility is an absolute priority. Like, imagine, for example, if we could use AI to finally unlock this promise of personalized learning. We still need a classroom. We need the teacher there because only a teacher can really provide the infrastructure in a classroom. But every student learns differently. I see this with my twins. You know, they were raised the same, and yet they're so different in how they learn and their ability to retain or understand information. So, one is dyslexic, and therefore the support and scaffolding that she needs is so different than my one who's like a traditional rote learner and can go down a rabbit hole.
The ability for generative AI to be able to take in text or video or voice and to translate it, and not just give an answer but actually break down the steps, has really transformed how they learn. So to me, this access is so critical because if we can really meet people where they're at and unleash their human potential, that could be transformative. How many people have we left behind in generations? The only way we're going to build a better future is if more people are part of it.
Yeah. And now there's so much concern. I mean, there's been concerns about sort of technology gaps and whether it was people having access to broadband or computers. And now it's a whole other risk around that if these gaps persist. Another thing you mentioned early on in this conversation was this idea of responsibility and how do you operationalize responsibility. And responsibility around AI was part of the founding mission of Google DeepMind. So what does that mean given that now there's such a heated conversation about the risks of AI and what kind of responsibility these leaders should be taking? Where does responsibility fit into your organization?
As you said, responsibility has been core to everything that has been at DeepMind from the very start. And when I joined, it was actually my motivation for joining. My thought was if I could bring that experience from building new technologies and new industries into this moment in AI, then that would be good. And in fact, it meant that my first day, the top priority was like, how do we operationalize all this ambition, which is really difficult because there's the technological aspect of it, there's the area of research on like ethics, psychology, child psychology, etc. So there's the research, and then there's the culture, and then also the governance.
So how do you have the right people around the room to decide not only what type of research to do, but, you know, is a model being developed correctly? Now, Google is an extraordinary organization that has a lot of trust and safety support, but our intent was always, can we do it from the very beginning? You don't bolt this on at the very end. You don't just have it as a compliance checklist. You need to be sitting and working hand in hand with the technical researchers so that when they're developing the models, we can think about what are the risks, how are we going to mitigate them, but also what are the opportunities and how do we maximize the opportunities.
A great example is AlphaFold. We did so much diligence from the very beginning to say what could go right and what could go wrong. What could go right is what led to the database that made it freely available to over 4 million researchers worldwide. What could go wrong led to us seeking external input, doing additional testing, not at the very end but throughout the process, to say whether this was worthy enough to release so broadly.
So when you say operationalize, what you really mean is building into every step of the process, having it be part of the culture or the fabric of the organization that if you're going to develop something from moment one, you think, is this something we should develop? What are the risks of this? And then you continue that conversation throughout.
Yes. And I think there's something about, from a leader's perspective, of just being present. This is a priority. I made the time to join every single governance meeting that we had. And the governance meetings were more about having those deliberations, and we were not all on the same page because, again, you've got the bioethicist with the neuroscientist and the policy expert having robust debates. But that's actually what we needed to really tease out of, should we be doing this type of research? How should we be structuring it? Who are the external parties that we should be working with to kind of check our work? Who should we partner with to make sure that it's having maximum benefit.
So there's something as a leader, which is, are you making the time and the focus and letting people know that this is important? I felt like we finally hit that point when I could step out of a meeting and I knew that those deliberations were happening because that then meant it was part of the culture. And, you know, again, we've hired, I know there's a kind of a movement around philosophers and ethicists. We've been hiring those for such a long time and having them as a core part of our organization because we really do believe this multidisciplinary approach enables a better product in the end.
There's been a lot of talk recently about the potential dangers of AI. Dario Amodei from Anthropic wrote a big essay that has sparked a much broader conversation, really involving everyone in the industry, wondering whether AI's development should be paced, and whether pacing means slowing down or does pacing just mean reallocating resources towards trust and safety. How should this be regulated? This has become a national conversation, this question of the risk of AI has become a national conversation. How do you respond to these headlines, signs and warnings about the potential risks of AI?
Yeah, I mean, like I said from the very beginning, I spent 50 hours interviewing for this role because I thought it's such a critical topic. This could go really right or really wrong, and how do we make, am I doing the right job by going into this and bringing that experience, and can I tuck my daughters in at night knowing what mom is working on? And I feel like I diligenced that upfront, which is why I've put so much focus on it. But it does mean that even to this very day, what keeps me up at night is, are we thinking about all the risks? Are we doing the best that we can? Are we responsible stewards with building and deploying this technology? And are we maximizing the opportunity?
We've done things like we helped to co-found the Frontier Model Forum, which brings together the different labs to work with the US government, for example. We've also published our Frontier Safety Framework, which outlines different critical levels that we think about and how we're testing to bring some transparency into the process. We show up and help to steer conversations around policy because we think this is such a transformational technology. It has to be regulated and it has to be done right-sized to be able to still have the opportunities.
In fact, Demis Hassabis, our co-founder and chair, published an article earlier this year in July outlining his vision for how we could regulate this technology with the US actually taking a lead in it. But as we all know, this is not about any one company or country. We actually need to work more globally and collaboratively in order to really make sure that we're developing and delivering AI in a way that is meaningful.
It sounds like you have from day one at this company felt like you were doing your part around trust and safety and responsible development of AI. But as you look at the fear that is now in the public, I mean, look, it was even on SNL, Saturday Night Live had a segment about fears that AI was going to kill the world or take over the world. These very big, almost, you know, impossible to fathom concerns as that's now in the mainstream. How do you address those fears? Do you think that the rest of the industry is doing its part? Do you feel reassured by seeing the president and a group of AI leaders, including Google's CEO, talk about what they're going to do to come together and tackle this?
I think it's so important that we're having these conversations. I don't mean to sound trite by any means, but, you know, maybe a couple of things on my mind. First of all, I was in venture capital during the rise of social media, and one thing that was really missing from a lot of the investment conversations was a talk about responsibility and safety. We're talking about this, which I think is fantastic, and we're talking about it across a wide variety of stakeholders, which is also important.
You know, I like to think that Hollywood has done us a favor, right? We have science fiction movies that provide us a common language and some common context of what could go wrong. And that means that we're having these conversations. In fact, I never would have guessed that around the dinner table I would be having conversations on what it means to be human. If you had told me that even two years ago, I would have been surprised by that. So I think it means that we're being more responsible because we're having these conversations. We don't have to agree right now, but we need to at least get on the same page of where we think the boundaries are and what are we going to do about it.
Again, I think it's really important this is a global conversation. When you release a model, it's available on everyone's devices immediately. It's not like during the computing days where we had to think about how do you, you know, the timeline to build all the components inside the computer, get the computer shipped, think about the internet connectivity, etc. We had plenty of time to plan. We don't have that type of time right now. So the conversations are important, action is equally important, and things like some of the global governance events, the AI Action Summits, have been ways for global conversation to continue to advance.
So it sounds like you're reassured and optimistic.
I am. I couldn't, you know, I would say that I like to think I'm very practical. There are risks, there are opportunities, we have to work together. So how are we going to go about doing that? I think of myself very much as a builder. And I don't think I could do this job if I didn't believe that the benefits will outweigh the risks. We just need to be thoughtful in how we build the future.
You've had an incredible career. You mentioned venture capital. You took this big role and left Silicon Valley and came to London. How did you get into this transformative role and defy all the odds, especially as a woman in tech? You were very rare to be a woman who was at the center of the technological revolution that is AI. How did you get here?
Yeah. When I look back, I've entered the stage of my life where I'm doing a lot of reflecting, and one of the things that I think a lot about is what set me on this path in the first place. So, you know, I'm an electrical engineer by training. I was in the semiconductor industry for 20 years, and it was really my dad. And my dad was an immigrant, so I'm a daughter of immigrants, but my dad is also an engineer. And my dad's origin story was that a teacher saw, he grew up in an orphanage, a teacher saw a spark in him, and it completely changed his career trajectory, and he made it to the US to study engineering, and that's what I saw when I was growing up.
And so to me, in the type of work he did, engineering was always about solving really complex problems to help people out. And I'm like, if an orphan boy from Lebanon can do this, well, surely I can. And then I went into my engineering classes, of course, and I looked around, I'm like, no one looks like me. So it has always been something of being able to see opportunity and potential where others might not, and being very true to my values. So if you ask me a question and I don't agree, I would say it regardless of where I was, but I would always do so with kindness.
And I think what that meant is I always took roles where I could learn with and from the people around me, but also have the space to pioneer and to do things that no one else was doing. Maybe that's what eventually led me to the series of decisions to make a bunch of right-hand turns along the way when it was so not obvious to anybody else, especially leaving Silicon Valley to come to London. So I think, you know, solid values, knowing what's important to me, what's going to give the most out of me, and really just the throughline of technology for making the world a better place and making sure that others like my dad have the opportunity to succeed as well.
Yeah, hearing your father's story, it totally makes sense why you're so passionate about accessibility. And it also sounds like being the only woman in an engineering classroom didn't discourage you.
