How Google DeepMind is building AI that can help humanity

0:00Starting a business comes with its share of ups and downs, which is why staying true to your vision is essential. a non-negotiable for Romeo and Milka Bregali, Capital One business customers and co-owners of Ra's plant-based restaurant in New York. Romeo and Milka took a leap of faith when starting their own restaurant, gutting an empty space and building it from the ground up. Every pipe, every wall, every detail. But building from scratch came with a heavy financial burden, which is when they turned to their Capital One business card. With the flexibility of the card's no preset spending limit, they were able to spend more and earn more rewards while bringing their vision to life.

0:36Today, Raz's success is proof that with passion and the right support, it's possible to make your dreams a reality. Learn more at CapitalOne.com slash business cards.

0:52When I was growing up, I saw my dad and he would come home from work. He was an electrical engineer. and he would set out these like beautiful pieces of paper with colored pencils and make gorgeous designs that would then turn into these microchips that would then go to power things like heart pacemakers. And I grew up with this, I would say almost like this curiosity of how could this young boy from Lebanon who was orphaned at the age of five end up designing technology that looked like art that would save millions of people's lives. And that's really how I ended up in engineering, was this combination of art, math, and science that could benefit people.

1:42That's Laila Ebrahim. She's now decades into an impressive career and currently leads a powerhouse AI lab. As chief operating officer at Google DeepMind, Laila remains focused on creating technology that benefits people, developing innovative projects at scale. You'll recognize some of DeepMind's household products, like Google's AI assistant, Gemini. But a lot of its work isn't consumer-facing. For example, they are the Nobel Prize-winning team behind AlphaFold, an AI tool that's revolutionizing scientific research. And today, Laila is peeling back the curtain on how the lab is working on the frontiers of AI in education and science.

2:24Plus, you might get one of the best everyday AI hacks I've heard in a while. I'm Rana El-Khalyubi, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

2:47Hi, Laila. Thank you so much for joining us on Pioneers of AI. Thank you. Excited to be here today. So before we dig into Google DeepMind, I want to go to your story and your background. So we're both Arab American and we have that in common. We're also kind of one of the few women still in a very male dominated field. So I would love to hear your story. Like, where did you grow up? How did you end up in tech? It's been quite a journey. So I actually am the first from both sides of my family to be born in the U.S. My father is Lebanese. My mother is Palestinian. they met in the U.S. Both had come for their education and ended up staying and eventually meeting.

3:29I grew up in Lafayette, Indiana, home of Purdue University. So it was quite an extraordinary place to grow up, to be around a university as part of my educational upbringing. It was fantastic. But when I was in elementary, middle, and high school, I was like the foreigner in my high school class. And so we kind of stuck out. I ended up going to Purdue University to study electrical engineering. And when I went into my first internship, it was with a company no one had heard of called Intel. It was quite ironic that I had this anti-computer mentality and then went on to go help design the Pentium microprocessor, which was the brains of the computer, which kind And it just goes to show you how much life can change.

4:18And now, of course, in the field of AI. Yeah. So before you joined Google DeepMind, you were at Kleiner Perkins. And Kleiner was one of Affectiva's early investors. And Mary Meeker was at Kleiner. And she was on the board of Affectiva for a couple of years. And she was one of the very first kind of early women in VC. And she paved the path for other women. And I just remember her in these board meetings. She and I were the only women around the table, and she held her own in a very strong and powerful way. How did that experience being in VC and also kind of overlapping with Mary Meeker shape what you do today?

4:58I think, as I just mentioned, growing up as kind of an outsider taught me to get more comfortable with myself from a very young age. And also, I think through the first 18 years at Intel, I really gained confidence in my ability to see things from a slightly different perspective. And that included things like when I moved to Japan in the 90s to work on this technology no one had heard of, DVD and USB. But I was willing to take that risk because I saw a lot of opportunity where maybe traditionally other people didn't. And my move to Kleiner Perkins into venture capital came immediately after Intel.

5:34I was recruited. And I realized I had been this entrepreneur my entire career of like new markets, new technologies. And what was it like to go work with entrepreneurs like you who have these big, crazy ideas and make things happen? I was very fortunate, again, there weren't that many women, but to have the opportunity to the queen of the internet, Mary Meeker, and hear the types of questions and how she thought about things, especially with all the experience she brought into that role. But it wasn't just her. It was the entrepreneurs, right? At the end of the day, these were and are people who have these crazy ideas and these ambitions and these visions of what might actually be possible in the world.

6:16And what I appreciated about that time in venture capital was seeing such a broad range of entrepreneurs. But what I really missed was building. And, you know, I'm an operator at heart. And so I actually went into one of our portfolio companies into Coursera, where I had a chance to take the experience of learning about what my fellow partners, the questions they were asking, how the entrepreneurs are thinking about building their companies, and apply some of that into an early-stage startup. It was about 40 people when I started. That company, Coursera, is now a leader in online learning with more than 117 million users and a market cap of$1.38 billion.

7:00I wanted to know how and why Lila made the jump from operating Coursera to Google DeepMind. Well, I, you know, I'm laughing because I was supposed to take a year off. So I had been at Coursera for a while and I thought, you know, I'm approaching my 50s and I really want to be thoughtful about this next chapter of my career. And I feel like I've had this extraordinary luck of being in the right roles at the right time and making cool things happen and having a big impact in the world. And so I was going to take a year off and really focus on my nonprofit. But John Doerr from my Kleiner Perkins days said, Lila, I just want you to meet this one entrepreneur.

7:46He sat on the board of Alphabet, and I eventually caved in and thought, I'll just do this one meeting as a favor to John. Little did I realize here I would be, you know, seven and a half years later. But I really wasn't sure because I didn't have a background in machine learning or artificial intelligence. And I was also based in Silicon Valley, and the role was in London. So I ended up spending 50 hours, 5-0 hours interviewing for this role before I decided it was the right role and I was the right person for this role. Wow, that's wild. You know, people often think about like these long interview processes as a negative, but actually, you know, if you turn it on its head, it's a way to ensure that this is the right next step and adventure for you.

8:35That's pretty cool. So what do you do as a COO? So my role has evolved quite a lot over the past seven and a half years. But I oversee all of the central operations of how we organize to deliver our research and our products into the market. But I also oversee all of our responsibility and frontier safety work and our external engagement work now, whether that's collaborating with policymakers around the future direction of AI to our work around impact acceleration for social good. So it's quite a broad remit. But what I like about a chief operating role, COO role, is really partnering with the organization to help achieve the mission.

9:18If I say, like, our mission is to build AI responsibly to benefit humanity, and I get to help shape the how we do this, it's really been an exciting, exciting journey so far. Yeah, absolutely. So we are living through this crazy technological shift, right? And some people are, you know, charging full steam ahead with AI. Some others are a little bit more skeptical. What's your framework for thinking about how we should be building AI and deploying it at scale in the world? So much has changed since we started on this journey. So DeepMind was founded in 2010 here in London, in 2014 acquired by Google.

10:00I think there's something here where we are constantly thinking about how do we be responsible stewards of the technology and how we develop, how we govern, how we roll it out. So we need to be thoughtful and responsible at the same time, making sure that while we're managing risk, also investing in the opportunity. Because at the end of the day, that's, I think, why many of us are here. It's not the technology for technology's sake. It's actually really wanting to make a positive impact on how people work, how they live, how they learn, on our understanding of the universe around us, helping to address some of society's biggest challenges.

10:38Yeah. Yeah. So let's talk about some of these and some of these examples. We have talked about AlphaFold before on the show, but for people who are not familiar, Can you recap what that is about and how are you making it available to other researchers to accelerate scientific discoveries? So AlphaFold is our advanced AI system that helps predict the 3D structure of a protein. And so if you think of things like Parkinson's, Alzheimer's, malaria, these are all protein-based diseases. And this is why proteins are important. And if you can understand how a protein folds, you can understand when it misfolds, what might be wrong.

11:17And that helps us understand diseases. And it also helps us deal with things like how to deal with industrial waste. Why are some crops more resilient to disease than others? There's all sorts of interesting things that we can really do in this space. So we developed an AI model specifically to predict the 3D structure of a protein. And actually, that's a very challenging problem to solve. Like before AI could figure it out, like we did not have a way to actually do that at scale, right? Right. At scale is key because it used to take a PhD student about four to five years with the right equipment, with the right experience to just do one protein prediction.

11:57There are 250 million known proteins. So like think about it as like a billion years of research, right? Now consolidated all within the past five years or so. And we did something that I feel very proud of, which was as we were thinking about how to actually release this, we sought outside experts to help complement and make sure that we weren't getting stuck in our own insider bias of was this safe to release, how should we release it. And the result of that led to a partnership with the European Bioinformatics Lab to publish everything in a database available freely to scientists worldwide.

12:36all 200 million plus protein. So one of the things we said was like, if we're going to give this to the world, let's make sure there's equitable access. So we've done some really interesting things. One is with the Neglected Disease Institute. What's an example of a neglected disease? So Leishmaniasis is an example of a protein-based disease that has actually impacted more people than COVID. It's just over a longer period of time. And because of that, it hasn't gotten the type of pharmaceutical funding that it might otherwise have gotten. So that would be one example. We also took a look at some of the usage from the database and realized that the continent of Africa had low usage.

13:16So we worked with the community across the many countries to say, how do we actually do a train-the-trainer to help with onboarding researchers so people who might not have otherwise have even had the access to the lab equipment can now advance some of the work in the fields that they're dealing with. So I think what's really exciting about AlphaFold as an example, a couple of things. One is that we went from model to impact in a very short period of time, and we're still quite early in it. So last year we got the Nobel Prize for our work in this space, two of my colleagues. Then the other thing that I think is quite significant about this is really saying, where does the human ingenuity come in?

13:56Because we may have had the model, But this is not about any one company, not any one country, not any one field. So the fact that it's being used so broadly, globally, over 2.9 million researchers are using this worldwide. I think it's quite extraordinary to think of what this might do to open up our understanding of so many fields around us. So powerful. And it changes. Yeah, it changes the way we do this kind of research. Now, you also do a lot of work around weather prediction and kind of mitigating the effects of climate change. Can you tell us some more about this work? One of the areas I'm really excited about is some of the work around weather prediction, because to me, it's like completely chaotic and unpredictable.

14:40But you live in the UK. You live in London. So that's part of it. I've got my brawly over here and my wellies there. Yeah, it's completely unpredictable. So, you know, one of the things that's been really exciting is some of our work around WeatherNext, which is our 15-day state-of-the-art forecasting. Again, we bring the AI expertise and we work with the scientists to apply it into their field. So we've worked with meteorologists worldwide to come up with a 15-day forecasting. We've also, working with the U.S. Hurricane Center on hurricane prediction. How can we predict 50 different potential paths that a hurricane might take?

15:18Like, and you can imagine what this means for emergency preparedness. And I think we're still, again, we're still in the early stages and like, imagining what might be possible in a few years and how can we avoid some of the crises that we've seen in our lifetime. So I want to put both of our like investor hats on because I liked what you said, right? Like Google DeepMind is developing the underlying AI that is going to unlock all these vertical applications of AI. Hi. With an investor hat on, where do you think the opportunity is for a startup versus kind of what DeepMind's doing? Like what makes a startup competitive, sustainable, have a moat given that you guys are building all these underlying key enabling technologies?

16:01I think a lot of the models now are available to developers to actually add in. They're like, what are the problems that need to be solved? And I think even when we did AlphaFold, we hadn't, you know, we didn't expect it to be used in agriculture in the way that it's being used. And it actually took me back to in the late 90s, early 2000s, when we were working on the computer and the internet build out. And there were all of these questions of like, oh, you know, our computers are going to displace the farmers or the teachers, when in fact what it did is it changed how people worked. It opened up opportunities that they hadn't even imagined.

16:34And I think we're at the early days of AI and we're going to see this, the creativity and the vision that entrepreneurs have of like solving problems sometimes that we didn't even realize were a problem because we're just so used to how things work. There really isn't much controversy when it comes to AI-enabled scientific discovery. Wouldn't it be awesome if we had more accurate weather prediction tools? But DeepMind is also working on AI applications that can revolutionize the way we learn. And when it comes to education, there are a lot of opinions about where AI fits in. We get to that in a minute after a short break.

17:14Stay with us.

17:25If you've spent any time building AI products or leading technical teams, you know this. Transformation doesn't fail because of ideas. It fails because teams can't move together. Enter Atlassian's Teamwork Collection. It has planning in Jira, documentation in Confluence, video updates in Loom, and now AI agents in Rovo, which connects the dots across your work so nothing gets lost. It's one AI-powered teamwork platform designed for how modern teams actually build. Learn more at Atlassian.com slash TeamChanger. That's A-T-L-A-S-S-I-A-N dot com slash TeamChanger.

18:14So let's talk about education because this is something we're both very passionate about. So you mentioned you were COO at Coursera for a while, and then you're spending more and more of your time at Google DeepMind now thinking about how to apply AI in a way that democratizes access to education. Why do you think AI could play a key role in education and how we learn? Well, I think education and learning, you know, it's in my roots from our heritage and I think even just the opportunities that it opened up along the way. And even back in 2000, I went to set up a computer lab at the orphanage my father was raised.

18:53In Lebanon? Yeah. There were a thousand students and, you know, they didn't have the same type of access to books and teachers. And, you know, subsequently continuing to build out the computer lab. Like you can imagine then of like students now having access to the same skills that some of the most advanced schools had in the world. And it actually spurred me to start a nonprofit called Team for Tech, where we work with nonprofits globally that are focused on providing ed tech solutions to help with learning outcomes. So education and, like, giving to the next generation so they have the opportunities is something that's just—it was a part of how I grew up.

19:38It is something that, like, I feel is, like, where and how I want to make my mark in the world and, like, leave it for my kids' generation and the ones that follow. And I feel really fortunate. When I came into this role, I wasn't sure where that intersection would be seven years ago. But I feel like where the models are at right now, they're getting factual enough. They're grounded enough. Like the ability to interact, Gemini was multimodal from the start. And what that means is you can now meet a learner with where they're at. Maybe they don't want to type in a long prompt. Maybe they want to talk, right?

20:16Maybe they want to take a picture instead. So that's kind of like we've been talking about personalized learning for so long. And I feel like we actually are at the cusp of being able to unlock some of this. You know, I want to go back to your kind of story with your parents and the role that education and learning has played in technology. And I feel like that really resonated because both my parents are technologists. They met at a COBOL programming class. My dad taught COBOL and my mom attended the class in Cairo in the 70s, right? And I feel like technology was my window to economic mobility, to, you know, the reason I'm in Boston now is because my parents really invested in our education and we were at the forefront of technology growing up.

21:03And it's kind of also my lens on investing. Like, I really want to make sure that AI is applied in a way that gives social and economic opportunity to people. So do you have any additional thoughts on that? Yeah, I mean, that's, I feel, I've been struggling to find the words because it's actually quite emotional for me. Like, I, when I took this role, I really felt like it was a, it may sound cheesy, but like a moral calling, right? I had no idea that the AI industry would shift so much in the past seven years. And so I sit here feeling incredibly fortunate and very humbled by the fact that I am in this role in this moment in time in AI's history.

21:51And all of a sudden, my very circuitous, weird background makes so much sense. if I can help to make sure that we're continuously thinking about how are we building AI with community so it happens with them and not to them. We as builders owe it to society, owe it to our past and to our future to be investing in this way. Let's get a little bit more tactical. I serve on my son's school board and - Oh, how is that going? I'm convinced that - Yeah, actually, so I started a couple of years ago. right after the chat GPT moment, right? And I think everybody realized like, oh my God, like we need to have a roadmap on how we're applying AI at school.

22:37And it's kind of interesting. A lot of educators are, their approach to AI is that it's cheating, right? And then separately, MIT just published a study essentially raising flags that over-reliance on AI could decrease your cognitive abilities and critical thinking abilities. What's your framework thinking about AI and education? Because I actually think we absolutely have to have our kids. My son is 16. I know you have twin daughters who are 15. I think it's amazing that my son is AI forward, and I want him to be using these technologies. But how do you use it in the right way that enhances learning as opposed to take away from your kind of learning experiences?

23:17I think that's exactly why we need to be encouraging the responsible use of the technology and developing the healthy habits from a younger age, the technology is not going away. So how do you onboard in a way that it's like getting clear about what is appropriate use? When are you using it for idea generation versus idea replacement? And I think back to when I was in school, people were questioning, was a calculator? What was that going to do to our math skills? Or you think about like photo editing and was that cheating on your photo, you're taking pictures. So as a society, we're not having the conversations that we need to be having in order to shape the technology in a way that it can be meaningful.

24:00I want to kind of dive deeper into that because I think that's a question a lot of parents and a lot of educators are grappling with. And I'll just draw from my personal experience. So Adam, my 16-year-old, he's doing some research project over the summer and he's been going to AI to ask about like summarizations of articles. And I actually had a conversation with them. I'm like, you know, this is great. It's actually helpful to get you starting to think about how to approach this research. But at the end of the day, you're gonna have to just read that paper. So we're kind of having these conversations on, what's the right use of AI and where do you use it in a way that, yeah, helps you get to the answer faster, but you're still, I don't know, you still have to do some of the basic work.

24:44Yes, and you know, the reality is you end up with some students who are not fast readers or dyslexic who may have challenges reading who might otherwise be left behind or start getting labeled as like, oh, you know, why aren't you keeping up with your classwork? And you're all of a sudden falling further and further behind. And I think this is, again, where this can make such a big difference. Like, where can the technology actually meet the learner with where they're at and help them on the learning journey so that like maybe in order for you to be good at algebra, you need to have your fractions.

25:17But maybe if you're not getting like one part of fractions right, like it's completely ruined your math trajectory. And so I think that's where I'm hopeful. Like I'm grounded in my hope, which is how can we use AI to really fulfill the human potential and do it in a way where it's not judgmental. So I've been thinking a lot about like tutors, right? A lot of AI companies are talking about in the education space, when you're talking with educators about like, imagine a personalized tutor for everyone. So I think there's some really interesting things that can happen with AI, and it doesn't replace the teacher.

25:54I really believe the human to human connection is so important. But in a classroom, then it frees up the teacher to actually do what the teacher has the magical ability to do, right, of like connecting with the students to helping them on their learning journeys. It just happens that a classroom of 38 kids, everyone's going to be on their own learning journey. Yeah. You know, you talked about Learn LM and these kind of tutoring modules. How do you ensure that these LMs are not hallucinating, that they're accurate if kids are going to be relying on them to get a lot of their information? I mean, this has been the big question with the large models, right, of like why it's so important that we think about how to responsibly develop them from the start and not like in our final testing before release.