Hello and welcome to the Tech Download, a new CNBC original podcast where we unpack the tech stories that matter most. Each season, we dive into one big theme and what it means for your money with insights from the industry's most influential voices.
I feel like it's our duty, our moral duty in this generation, to make sure that we're providing the infrastructure for the students and the teachers to help shape what the future looks like.
This season, we're looking at Google DeepMind, the powerhouse driving the tech giant's AI push. We've been given rare access to key figures at the company. In this second episode, we speak to DeepMind COO Lila Ibrahim about how AI could impact global education and the way we live.
There's something here that we know there will be a disruption. So, how do we approach this as responsible stewards of the technology?
Hey everyone, I'm Arjun Kharpal in London and Steve Kovach here in New York. This is the second episode of our three parts on Google DeepMind. And in the last episode, we heard from the CEO and co-founder of Google DeepMind, Demis Hassabis, talking about everything from the next frontier of AI development to China to bubbles and to how this company works with the broader Google business. In this episode, we're going to hear from the COO of Google DeepMind, Lila Ibrahim. But first, Steve, I want to ask you something because it's something that all our listeners, all our viewers can relate to. How are you using AI right now?
Yeah, so I've been using Gemini for the last couple of months since Gemini 3 came out. Just I like to test everything. So, I've been kind of on a Gemini kick for the last couple of months. And you know, one thing that I loved about your conversation with Demis was how he is a gamer and how that has kind of informed a lot of his product development and his career. I know you're a gamer as well. I'm a gamer. You would almost think we're a bunch of millennial men working in this industry here. But one way I've been using Gemini recently is I started playing this game called Expedition 33. I'm sure a lot of our listeners are familiar with it. It won game of the year last year and I got stuck early on because it's such a trippy game. And so I thought, what if I just asked Gemini for help? So I literally pointed my camera at the TV, fired up Gemini and said, what game am I playing and can you help me? And within seconds, oh, you're playing Expedition 33. You're at this part on this level, and you need to go here. It just immediately gave me the answer. Literally sat there with me almost like a coach. And that was like one of those crystallizing moments in using AI that I was like, wow, this really works. Someone just needs to put a better layer on top of the product side of these things for people to understand all the capabilities that these large language models have. They just haven't been necessarily unlocked yet. You got to get kind of creative. Like I do other things too for productivity and work. I have, for example, I set Gemini up to scan twice a day, every day for SEC filings related to the public companies I follow and it works. I don't have to go there and check it myself. And you know, once upon a time, yes, that was possible. I would have had to code my own little web crawling bot to do that. Now I just say, tell the AI chatbot, do it for me. And it does. It's pretty cool stuff.
Yeah, that's so interesting. I'm always discovering new things to do with these chatbots, with AI. And I think that I'm going to kind of make a bit of a stretch to my next point, but it's all about education. It's all about learning. We're learning how to use these tools. We're learning how to do new things with some of these AI tools. I think it speaks to the way in which these leading AI companies believe, genuinely believe that AI is going to kind of be everywhere. And one of those areas is in the classroom, in schools, and in education. And that's the discussion I had with Lila Ibrahim, who is the Chief Operating Officer of Google DeepMind.
Lila, you and I last spoke in I think January 2024. It was at Davos. At the end of that conversation, you said to me, you're really excited about the opportunities in AI and education. So, you know, we're a couple years on now. In that time, what has changed and what's got you excited?
So much has changed in a very short period of time, hasn't it? You know, when I maybe just a quick reflection on the journey of AI and education over the past few years. You know, within Google DeepMind, we tend to focus on these big scientific problems. We're founded by scientists and it's really in the ethos of how we operate and what we do and how we approach things. And a few years ago, about three years ago, we decided, let's start treating learning like a first-class science problem. How do people learn? And if we could unlock that understanding, what might that mean for being able to help all different types of learners in all stages of life approach their learning differently, better? And so we worked with a lot of experts around the field of learning and education, pedagogical sciences, academics, thought leaders and took those learnings and fine-tuned a model which we called LearnLM. We have spent this year really infusing Gemini with LearnLM, all the features and all of the best learning because our thought was why should it be limited to one specific model? What if we could make our general purpose model have these same capabilities? And the way that AI has advanced over the past few years, we felt this was the year where we were ready to do that merge. So, it's been quite extraordinary because now Gemini has all those benefits that we've built in over the years and tested along with people worldwide.
How did that sort of come about? At what point were you thinking actually, you know what, we think that this model's good enough. You think we should be infused into, you know, the main Gemini product?
Yeah, I would say in Q1 I'm meeting with Demis, a very memorable meeting where we were coming in to talk about education and learning but did not know that we were calling now the time. And you know, Demis is quite extraordinary in his ability to think about timing, you know, I like to call it DeepMind time. Things can you can be working on something for quite a while, but then when the moment is just right, we go into a mode where it's like, okay, all hands on deck. What do we need to do to make this happen? And so what was exciting about right now too is it wasn't just about infusing all of our research approach around LearnLM into Gemini, but it was also putting that into the Gemini app through a feature called Guided Learning which isn't about giving you the answer, it's about taking you through the steps in the learning process. And I think the multimodality of Gemini also we felt we had were making enough advancements not only where we were at the beginning of this year but then our outlook of where the technology is heading, you know, most obviously with recent announcement of Gemini 3. So we felt like now was the time to really unlock this and then we can start focusing on how do we work with academic institutions, with governments, with consumers as well on bringing these learning features out more broadly and across the globe.
How do you infuse this into educational systems which let's be honest haven't been updating in such a long time and are still very much teaching the same curriculums and are operating in the same way? How do you bring it to say, hey, we have this AI tool. We have this product. We think education systems globally can benefit from that. And so how does this start to actually get into the classroom, I guess?
Yeah. And we do have, I think, the kind of this multi-prong approach of being able to make it available widely to consumers and people regardless of age or where they are in their career or learning academic environment. But also, you know, the benefit of being part of Google is Google has a very long history of providing tools and technologies and products into the classroom and working with academic institutions. So in many ways this is kind of an evolution of the features but we wanted to do it very thoughtfully. So we looked for different organizations where they were open to partnering. One example is in Northern Ireland. I was there earlier this year and you know, they were facing a real crisis with the teachers and teacher burnout which is quite understandable when you have a class of 30 plus students and you're trying to meet the needs of learners at different learning abilities and teaching a national curriculum. And what the government decided to do was really focus on how could they help with teacher productivity. So, they did a pilot over a six-month period of time and found the teachers actually reported back to me that they saved on average 10 hours per week per teacher. And they had written out post-it notes of like I got my evenings back with my family or I can now co-develop a curriculum for our class and think about how to bring all of my students along. So there's something about this localization of how do we bring the technology but we really need the expertise of the local leaders to think about how to bring this responsibly collaboratively into the environment. Now that's an existing infrastructure but to your point like we should be having conversations about, you know, do we just layer some of these tools and technologies in the existing with, you know, education system which was built for very different purpose. Actually in November we held an AI for learning forum where we brought together thought leaders from around the world to have a very candid and open conversation of how do we approach the future where what are the challenges and what are the opportunities because we may we have the technology expertise but we're not academic experts. We may have a lot of experience but this really does take a collaborative approach. So we started to explore the future and say where can we layer and where do we really need to rethink the system both in terms of teacher trainings, tools in the classrooms, how do you assess, what are you assessing for. I think these are all big questions.
A lot of these systems need overhauling for an age where AI could have a huge impact across the board in so many different industries on society more broadly. These tools are a part of it in terms of using them to learn but also learning how to live with and use AI as well. So as you think through education AI, how much of thinking is about, you know, talking to governments saying, well, we think maybe, you know, there might need to be a deeper conversation about bigger changes to education systems more broadly.
There is a lot of experimentation going on and you find different countries taking different approaches. A lot of times you'll see people say, okay, we know we're leaving students behind because what the jobs that we're training them for are going to change or the learning differences are more and more apparent. And so we're starting to do more of that co-definition of like what might a pilot look like with a slightly different approach. It's still early days with how quickly the capabilities are adapting. We can't leave these students behind. They need to learn how to use the technology responsibly. And I feel like it's our duty, our moral duty in this generation to make sure that we're providing the infrastructure for the students and the teachers to help shape what the future looks like.
Already, you know, there's a lot of discussion around AI tools being used to cheat papers or to write essays and things like that. How do you sort of address those kind of challenges? Is there sort of an AI-proof way of creating courses or assessments that you think, you know, may need to be developed as well?
Yeah, and I think if we only manage for the risks, we also miss out on the opportunities. So I think we need to be looking at these two together. We are doing some efficacy studies right now in the UK, in Africa, US, India where we're also working with the teachers to say like let's actually see what happens when the teacher actually integrates AI and role models the responsible use of it because if you can kind of teach the responsible use at an early age rather than students trying to hide that they're using AI, like is there a way that we can bring that cultural aspect into the classroom as well and I think that's going to be really important. When you look at past technology trends even I remember, you know, part of my roles in the past have been bringing computers into classroom and we had to really stress the fact that like listen, the computer's not the magic, the teachers are. So how do we first start with a teacher because a teacher knows the classroom and the curriculum. How do we think about when the technology can be used and should be used. Can we have consistency even within a school or within like because a lot of teachers will might approach it differently and can we think about how do we show where the opportunities are. So an example that I like to share is that I have twin daughters, right? So I'm running my A/B test every single day on how my kids learn. They're teenagers. They have very different ways of learning. And one of my daughters is dyslexic. And the traditional school system is not designed for her at all. And so I have seen how AI has actually unlocked her potential and shifted her from being having a lack of confidence of labeling herself as not smart and instead being able to do things that she never thought possible like representing her school as a communications lead for student council. She's able to now put her thoughts together in a way and communicate it such that others can understand. And I think this is really where AI can shine. It's like we need to focus on the risk, but can we stop just always labeling it bad and think about actually how can we put it to use responsibly to unlock the human potential and then how might people flourish based on that.
So Steve, there's this debate right now particularly amongst people who are in high school about whether they want to go to university or college or do what we call here an apprenticeship, getting a job but studying at the same time and getting paid as well. So this is a route that's become incredibly popular with people who now don't see university necessarily as the best route. And I think there's all this kind of talk around in the age of AI, what kind of skills are going to be important where AI can do and automate so many of the tasks perhaps that we would have gone to university for to study.
So what we're seeing more here in the labor market is entry level jobs. So it's actually people getting out of college trying to find those white collar jobs. Perhaps we can learn something from our friends across the pond because when these folks are graduating with, you know, university degrees trying to work in corporate America, we're in this kind of low hire, low fire, to quote our Fed chair, situation here in the labor market. And some people are blaming artificial intelligence for that. I think it's probably more uncertainty around tariffs, around the economy. However, you can see that kind of mindset taking place as AI takes up more of these computer desktop jobs.
I think there's a multitude of factors now. I think there's this view that the value of certain degrees has gone down, but I also think it's genuinely, you know, there's just more options now. For a lot of people, higher education isn't necessarily something that suits everyone. It was the traditional route but I think those traditional routes are breaking down now and I do think part of it is the way in which people are seeing technology develop. I think there is slowly becoming this realization that actually, what do I do in a world where AI can do so many of these jobs or so many of these processes? Where can I best train and use my skills? And I think Lila made an interesting point about what this means in education from a young age. That they need to use AI tools and learn how to use them or risk getting left behind. And so I think that's quite a pertinent view.
She was really interesting the way she described it too. It's not just about educating people about AI. It's like these are the tools to your point that you're going to need to be using, right? That these tools are whether you like it or not this is how the digital world is going to operate when these kids grow up and so teach them now in order to be prepared for that. Obviously then the question becomes what about people our age, you know, in the 30s and 40s and 50s who are mid-career or late career, how do they adapt to that? That's going to be a bigger challenge.
There's lots of things to think about as you roll this out but one of those is what is going to be the impact on students, on teachers, and you know, there's questions over whether this kind of technology will start to erode things like critical thinking and the ability for us to think in that way. How are you tackling these kind of questions?
Maybe if I zoom out first. When Demis started building DeepMind, they took a very interdisciplinary approach to AI which was quite unusual but what that means is every time we approach a research project we think about what are the right skills around the table and we tend to have people from a variety of backgrounds. So the person who led our LearnLM technical work, Irina, actually has a psychology background and then went into computational neuroscience and AI for her PhD. We tend to have these people with these very diverse backgrounds. Some child psychologists, we have ethicists. So everything from the research definition to how we do our governance to how we think about the external partners that we need to be bringing their voices in because regardless of what type of team we build, we're still not we want to avoid insider bias. In fact, one of the areas I'm particularly proud of is when we released our first technical paper around some of the large language models. We did it in collaboration with our ethics research team and similarly when we did our agent technology, our Astra Demo, we released a paper on the ethics of advanced AI assistant. So I think this is just kind of the part of the DNA of how we work. Hire great people who ask big questions. Make sure the team is diverse in terms of expertise and they'll ask better questions. Take a collaborative approach. Bring the outside voices in from the very beginning. Make sure that responsibility is woven through all teams. So by the time you're releasing a model, it's already built in. You're not quickly trying to fix things.
That's really interesting. And that's also a question over making sure this technology is getting into the hands of as many people as possible regardless of background, where they live, and this kind of idea of equitable access to the technology as well. And so as you think through those kind of opportunities there, what are some of the things you know you could do at DeepMind to improve access to this?
Yeah. And this is, you know, been a passion of mine for about 25 years now. When I first built a computer lab at the orphanage where my father was raised, where, you know, they didn't have electricity. They...
