Hi, I'm Miriam Vogel with Equal AI, and we are so pleased to have you join us for this edition of An AI We Trust.
Welcome back to An AI We Trust. Today we are joined by two special guests. Our first guest holds a job title that did not exist until this year, and she thinks every AI company will need one. Lila Ibrahim is the Chief AI Readiness Officer at Google DeepMind, where she works with governments, educators, and communities to prepare the world for rapidly advancing artificial intelligence. It's a role she stepped into in January after nearly eight years as Google DeepMind's first-ever Chief Operating Officer, guiding the lab through some of the most consequential breakthroughs in modern science, including AlphaFold, which helped earn DeepMind's scientists a share of the Nobel Prize. Before Google DeepMind, she spent nearly two decades at Intel and then moved into venture capital at Kleiner Perkins and led Coursera as president and chief operating officer, bringing online learning to millions. She co-founded the education nonprofit Team for Tech and sits on the UK's AI Council. She has been named not once but twice to the Time 100 AI list. We are particularly excited to hear about her current translation work, something we deeply care about at Equal AI, where she builds the bridge between the people creating AI and those who live with it from inside one of the world's most powerful AI labs. We are also thrilled to be joined by her colleague at Google DeepMind, Lucy Calcott. Lucy works with Lila and builds internal deployment tools, including scenario planning models. She helps partners, governments, and education institutions map the consequences of strategy choices and decisions across systems and planning under uncertainty, which is certainly something we all need to be learning from. Lucy has led initiatives and projects all over the world, from developing early childhood education in South Africa and leading public policy programs in Haiti to supporting an India-based organization on microfinancing for female entrepreneurs and small business owners. We understand she is known as a team AI guru, hosts hackathons, and offers fresh perspectives. Lila, Lucy, thank you so much for joining us.
Thank you, Miriam. It's a pleasure to be here with you today.
Thank you, Lila. Let me start with you. Your path at Google DeepMind, starting from being an electrical engineer in prior lives to COO to Chief AI Readiness Officer, is certainly not a path someone would say is conventional. What do you think about your background in those different roles? Give you a unique perspective to take on this job, and what do you think often gets lost when technologists and policymakers and CEOs are trying to have a conversation about our AI future that is our present?
Oh, you know, I'm probably at that stage in my career now where I can look back and it makes way more sense to me than it did when I was at any point in that journey. And I can see the through line even from the early days of choosing to go into engineering. It was really about how do I make the world a better place? And to me, technology was the way to do that. And it was largely modeled after watching my father, an electrical engineer, do amazing artwork that just happened to be heart pacemakers and safety devices for cars and airplanes and other transportation devices. So I think I always grew up with this concept that technology is there to solve problems, to make the world a better place for human beings. And you know, I think there's another thing, which is throughout my career, which is I've always, because I've tended to be an outsider, I may have been an immigrant growing up in Indiana or one of the few female in my electrical engineering courses or working on a technology that didn't exist because nobody else of my colleagues wanted to jump at the opportunity. And I think what that did is it meant that I saw problems from a slightly different perspective, and there was never an option. I had to figure out how to solve it, and I could never solve it on my own because I didn't necessarily have the expertise. So a lot of it was how do you approach problems with curiosity and going across geographical boundaries as well. And but really thinking about like what problem am I trying to solve for humans and for people in general. And this is a through line throughout my entire career. And it happened to bring me to an organization with no AI when I had no AI background, right? So technology may be a through line. Impact and public benefit is a through line. And then in 2018, I had an opportunity and spent 50/50 hours interviewing for this role with DeepMind because why would I move to the UK to work on artificial intelligence when I was coming from Silicon Valley and AI meant a very different thing back just nearly eight years ago. And I think, but I think it's because I've had that unusual background, a wide variety of sectors and positions, being an outsider, learning to ask questions that actually prepared me very well for this moment in AI. Because, and you mentioned my coming into this new role, and part of that was I've spent so much of my career focused on the technology in of itself, but we're at a point with artificial intelligence where it's getting more general, it's getting more capable. We have to think about what's the societal impact and how do we ensure that AI gets developed with and for humanity and doesn't just happen to us. And the only way to do that is to really lean into this readiness concept of meeting communities where they're at and making sure that we're bringing them onto this journey and that we're all learning together because it also holds us to better account as model developers, ensuring that the AI that we're developing is meeting the greatest need for society.
That's so helpful, Lila. And something we are often talking about, that we have to think about who's developing the AI and the plans for deployment. And your background, your personal experience, it sounds like it brings so much to your perspective and your work, which makes perfect sense, but just rings home how important it is. Who is leading in AI is pivotal, and the mindsets, the perspectives that we can bring to it are of vital importance, and the more the better. So to that point, Lucy, you've been focused on global affairs initiatives around the world, and that might be less typical for those in the AI engineering space, and I imagine brings you a unique perspective that you bring to your work. Can you tell us a little bit more about your background and how that helps facilitate your current role?
Yeah, I think similarly to what Lila said, I have had much less of a storied career, much less of a career in general. I'm very young to be working in an industry that's moving so fast. But I think broadly speaking, the journey is mostly being categorized by learning and trying to find problems to solve that I really cared about. AI is the most interesting technology that humans have ever built and developed that has the greatest potential for changing our lives in really useful ways, in positive ways. The background and the reasons why I got led towards working on global affairs and nonprofits and past programs that I helped run was caring about the people who would be impacted by either having access or not having access to really transformative technologies. And we've already seen that with past waves of technology. The internet was one, and I was very young when that was happening, and Lila lived through it and kind of helped create the access to parts of the world like the one that I grew up in. And I think we're kind of at another inflection point. And so the reasons why I ended up joining DeepMind were very much the same reasons as being interested in global affairs, which is how do you tackle these really big, sticky challenges in our world? And technology is, in many ways, this technology is the fastest route to doing that, and possibly could be a solution to problems that we've never been able to tackle, even through like really cool international cooperation agreements and policies in the past. So that's sort of the progression of thought that led me to what I'm doing now.
Very interesting. And you get to that point that we often hear about how AI will solve so many of our problems. You are seeing it firsthand. And I'm wondering how that led to your current role. Lila, why did you think, why did you all decide that you needed to have this role of Chief AI Readiness Officer? What does readiness look like? What does it mean? And why does it need to be a separate perspective?
I'll start with the last question first, which is it has to be a separate conversation because we need to be. The way I think about it is I am a leader in this moment in time in history. What is my responsibility, my accountability to my organization, to my organization, the team, the mission, and to society at large? And I think, and readiness doesn't mean one thing. It's not just about readying your employees. It's not just readying your processes in your organization. It's also your impact in the world and bringing those learnings back in. So, you know, there was no question that it should be a separate and that it shouldn't be nested under a specific job function, but something in and of itself, at least in this stage. And again, I think I'm the first in the world, and if I do it right, then other companies are going to be thinking about this as well. And readiness can mean so many different things. And in fact, one of the things I really worry about, and it keeps, there's a couple things that keep me up at night, and one of them is as AI continues to progress, will I have done, will we, will I, will we have done enough both to maximize the opportunities and manage the risks? But what I do know is that by having it as a separate, I have the accountability and we have the accountability collectively to focus on benefits and risks. And that could be everything with like you have to first start with access. Are people having access to the technology? Are they literate? Just because you have access doesn't mean you know how to use it. And I think a lot of the work that you're doing at Equal AI and the literacy programs are so critical here. But then are we thinking about productivity, which is I think everyone is discussing productivity, but are we paying enough attention to what are the uniquely human skills when we're using the technology that we want to keep, that we want to hire for, we want to develop, we want to retain, we want to reward? That's really critical. I don't think we're spending enough time having those conversations as well. Whether it's are we rewarding teachers for the test results? Are we rewarding teachers and school administrators for producing students who are ready for the workforce? Are we rewarding employees because they're just using the tools and they're more productive? Or are we rewarding them because they have brought their unique human skills and nuanced understanding to solve a problem more creatively, better for more people? So I think there's a lot of things with readiness that can be wrapped up into this. And what we're trying to do now is work with policymakers because governments have large constituents and can really have an impact there of like what are the risks we need to manage and how are we benefiting society at large. So that is one of the reasons we're doing a lot of the work with policymakers, and through those collaborations, we learn together. We learn so that we can shape the future, but we can also learn, like, have we mitigated the risks enough? Okay. When we did some of our work around some of the scientific tools, we realized that certain parts of the world were not using those tools because they had never, they didn't have access to previous examples or literacy. They didn't have the literacy. So we had to develop training programs. So the only way that we're going to build a better future, bring more people into it, address the problems quickly, and then move forward, and how do we co-imagine and co-create a better future for everyone.
Well, thank you for doing that. And I would love either of you if you could help us break down for our listeners. I think when you and I talk about AI literacy, we have an understanding of what we want that to look like, what we want people to understand, and the agency we want them to feel. But can you tell us when you're doing this work with governments, with educational institutions, what is the literacy that you're trying to help them achieve and understand? What does it look like if they were to be as AI literate as you would hope?
Maybe what we'll do is we'll jump into an example first. I'll pass it to Lucy, who's actually been doing some of these programs, including going, she was recently at Purdue University in Indiana doing some work on how do we help build AI literacy there. Maybe you want to give some specific examples.
Yeah, I think literacy is an area that is very different for AI than it has been for past technologies because it's such an infrastructure technology. And so it's not just understanding the technology itself, which is important, and there are lots of programs that DeepMind has worked really hard on, like Experience AI as an example, which kind of takes you through what is AI and how does it work and how is it built, but it's AI in something. And actually, like you were saying, the kind of last mile of actually deploying something and seeing it in action against a problem, solve a problem for you, is the hardest part. And so literacy is not just a composite of, okay, well, this is what the technology is and this is how it works, but how does it function in these different careers, professions, institutions, and especially as what roles are and how people work is shifting, what does it look like if you want to use it as an enablement tool? And so rather than thinking about it as like, okay, cool, I know how like differential equations work, like what is the application of AI in my worlds, in my life, and the problems that I want to solve? And so one of the things that we were looking at with Purdue was across the board of faculty, there were a whole bunch of different aspirations and interests in research areas. What does AI in that research question look like, and how can it really accelerate progress? Which is a different question to ask from just sort of understanding the overall title line of how AI works or how it functions. And I think breaking that down into a few different pieces would be like, one, what is AI? Two, what are the like elements of responsibility that you really need to consider if you're using it? How do you make sure that if you're trying to implement a solution, AI is not perfect? And we also want humans to be, as said, elevated by AI rather than just replaced by AI. And so there's a whole bunch of thinking that you need to do within literacy about where do you want to use AI and where don't you? What are the pieces of like fundamental human taste and judgment that you want to preserve? And then how do you know how to leverage that? So how do you bring your personal experience and the experience of some of these professors who had really long careers solving these problems into using AI in a way that really accelerated their research? So those are kind of three buckets that I was thinking about.
Well, I think it's really important to underscore part of the message you just offered, which is there are ways that we should not be using AI. And I think it's really important that at one of the world's leading labs that has won a Nobel for your discoveries in this space, you are saying there's ways to use it. There's ways we should not be using it. And in my work with you over the years, I've so benefited from understanding the depth of your thought on where, how that plays out internally, the training that you offer internally for all of your staff to be able to answer those questions. I think there's so many things that you model internally that I'd love for us to talk about today. So maybe let's start with the ways that you advise people how to use it, how not to use it. You know, when you have a breakthrough that's ready to leave the lab, I can imagine the enthusiasm and excitement about the benefit that you're going to offer the world. So how are you deciding how and when it's released? Who is asking the questions? What at a high level is the process to decide if and when it will go out the door?
It's been a real journey over the years. So when I started as chief operating officer, my initial focus was how do we establish what does it mean to be responsible stewards of this technology within a research environment. And you just highlighted one of the biggest challenges, which is you never wait until the very end to say, are you ready for the release? Have we been responsible stewards? You have to do it from the very beginning of when you're starting to scope what the problem, what the model's going to look like, or what the features are that you want to develop. And therefore, throughout every testing iteration, every building iteration, you're building this as just part of the DNA in which you're managing the development. So the way I like to think about it is, are we being responsible in our research? Are we being responsible in our development and responsible in our deployment? On the research side, we oftentimes have to think about what's around not just the next corner, but over multiple horizons and multiple corners. So an example of this would be some of the early work we did around persuasion, or before we actually even did the technology demonstration on Astra, our assistant technology, our universal, we published a paper on the ethics of advanced AI assistants. So we did, we have teams of social scientists that sit alongside our researcher, technical researchers, and ask the questions of what's coming, so that while we're developing it and to coincide with a technical development, we can think about these questions, develop the types of policies and evaluations to ensure that we're testing it. During the development stage, we are also, and deployment stage, we're also thinking about how do we test the technology. When we developed AlphaFold, our, as you mentioned, our Nobel-winning technology, we weren't sure if we had all the right voices around the table. So we actively went out and sought inputs from other bioethicists, Nobel laureates in the space of biology and chemistry. We asked different institutions and organizations like the European Bioinformatics Lab, is this tech, can you help us test this technology? Is it safe? What's the best way to deploy it? And that led to a database which now has nearly four million researchers worldwide using this to advance the fields of agriculture, understanding diseases for better therapies, how to deal with industrial waste. But that's only because we developed it in a way and deployed it in a way that the industry could be ready to do something with it. So we work a lot with organizations, with governments around the safety and security institutes, and we try to say this is not about any one company. It's not about any one country. We have to take, when a model is released, we have to make sure it's as good as it can be, knowing that it may not also be perfect, and therefore we have to be able to respond and react because we try to take a very much a learning mindset to everything that we're doing. When you're working in pioneering technology, you can't, as much as work you can do in advance is never going to be enough. So you have to be able to have the right checks and balances along the way. That's been a big cultural journey. Like if I think about AI's development over the past eight years, there was a time when the researchers couldn't even get their head wrapped around what the downstream deployment might look like. And it's why that very interdisciplinary approach is so important. We have to have people that are not only the machine learning experts, the people that are policy experts, but also the social scientists, the communications, that you need to have a broad group of people. And even that internal diversity is not enough, which is why it's important to have external voices. So I don't think, you know, everyone thinks that there's like some silver bullet. There isn't. This is really hard work. It's sometimes not sexy, but it's the most important work I think that you can do. And again, I think it takes that very senior level accountability. Like I showed up to more meetings than not because it was important to have the senior leader there who's going to take accountability and create the space for the robust conversations, and to the point where I could be out of the room and feel that those conversations could still continue. So I do think there's both a technical, human, process, but also you need to think about those as well as how do you build the culture to continuously improve because the technology is moving so quickly. What you tested for three, six months ago can easily be outdated.
There's so much you're sharing, Lila, that I hope everyone is breaking down and digesting in full. You know, you mentioned the culture. I've noticed the culture at Google DeepMind supports the ingenuity and the enthusiasm, but also the ethics and the moral impact of what you're putting into the world and making sure that it is safe for everyone's use. Making sure that you bring in the multistakeholder perspective. I love that you said there's no silver bullet. There's no AI tool that will create perfect AI for you, that it will always have new use cases, that will always drift in different directions depending on how you're using it and new consequences. I'm thinking about agentic AI will bring and flow in different directions that may or may not be expected. I love the curiosity and the humility that you're demonstrating and how you think about this because if anyone is going to be confident about AI use, it's going to be you and Google DeepMind, but that you are indicating and demonstrating that you have to be humble because there's a lot of uncertainty in your work, and I appreciate you telling us your perspective. You have to be looking many horizons into the future to understand how to do your job today. But there's also so much happening at this moment. It's a remarkable time for this conversation in particular. We're in the middle of the UN General Assembly week, and there's a session on AI risk and international coordination. The US and China are reportedly in discussions even today as Presidents Trump and Xi meet in a summit to talk about an emergency AI hotline modeled on the Cold War era red phone. And we're talking about how AI risk really could be bringing superpowers together, hopefully to have this conversation about how we do at a top line this work safely and aligned. Obviously, AI does not have borders, and so thinking about this as a geographical issue really is not sensical in so many ways. So I'm curious how this impacts your work while you're looking at the long horizon. There's so much that you're responding to, I'm sure, on the daily level given the headlines of the day and the speed of this conversation. So I'd love to hear how the last week's headlines in particular impact your work. What does that look like? How does that, if at all, impact your meetings, your focus, your approach to creating AI readiness?
Well, DeepMind was, just a quick history for your listeners, is that DeepMind was founded in 2010 here in London and acquired by Google in 2014, and then Google DeepMind evolved from kind of pulling together a range of AI and machine learning teams across Google in 2023. But the important thing to hear are two points. One is that since DeepMind's founding in 2010, risk has been at the very top of our thinking. You can't talk about one opportunity without talking about risk. It's baked into the values, the culture, who we hire, as you've heard, how we operate. And that's been really critical because the two technical founders have been thinking about this for a very long time. And you may have seen recently that Demis Hassabis published a paper on how to think about governance of advanced AI and artificial general intelligence, a term that was actually coined by his co-founder Shane Legg back when. So I think this has been just part of our DNA. It's something we've been knowing that the time would come and we've been working towards. And I think Demis' recent post about how do we have some global cooperation, I think the US is well positioned for this. So I think it's quite timely for the UN conversations. The second thing is being part of Google is a really interesting balance to that. The way I often describe it is like a very human one, which is, you know, my kid doesn't feel well, what's the first thing I do is I go do a Google search. And so, or like my mom falls and what do I do? I go do a Google search. So that trust that we have on a global level is something that we take very seriously. So I think that, you know, it's kind of a balance of both the DeepMind thinking about AI risks and benefits for a long time and being part of broader Google where this is taken very, very seriously. I also think it's quite wonderful that science fiction and movies have given us a common language to talk about what the risk might be. It's unlike my time in venture capital and the rise of social media. We didn't talk about risks as much as I spend talking about risks in my time in AI. The fact that we're talking about what does it mean to be human. Also, I think, you know, if we all went back in a time machine, I think we would have surprised ourselves by saying, okay, we're discussing this at both dinner parties to the UN General Assembly. People are having conversations about this topic. So I think these are really important because we would not be responsible if we weren't having these conversations. AI is such a powerful technology. It requires exceptional care and it requires this type of attention to think about how are we going to govern it. And regardless of where you fall on the risk spectrum, it's important that we engage different perspectives into this conversation and also take action in a collaborative and timely way.
