How AI Is Solving Problems We Thought Were Impossible | Yossi Matias, Google VP of Research

0:00Hello, everyone. Welcome to Silicon Valley Girl. I have an amazing guest today, and we're going to talk about AI and what's going on in the world. And I think you're the best people to talk to. Could you please introduce yourself? And for anyone who doesn't know you, why should they be listening to you when it comes to AI in our future?

0:16Yossi Matias:Well, thanks for having me. It's a pleasure to be here and talk about AI and the amazing things we can do with that. My name is Yossi Matthias. I'm head of Google Research. And I'm looking into how to advance research and AI for the benefit of so many amazing opportunities. And you launched one of the tools I absolutely love using called Google Trends. It's where you go and see what billions of people are searching for. And you also see how they search for these terms based on the season, for example. because I'm in part of my business language and sometimes people want to study more. And in August, we see that they don't want to study at all like December.

0:54So I really like that tool. So before we go into details of research, based on purely trends that you're seeing, what is the thing that you're seeing already, what we're still underestimating, but it's going to affect our lives in the year's future?

1:10Yossi Matias:First, we're perhaps in the most exciting time ever because we're seeing, in a way a dream come true of how to take and advance research in a way that impacts reality in a pace that is faster than ever. One thing I like to call the magic cycle of research is looking into how to drive research breakthroughs that matter. I like to call it as intentional curiosity. And then taking that research result and applying it back to reality and going this circle And something that as somebody who's been both a researcher, scientist all my life, and also a product guy, Google Trends was my first project at Google, actually.

1:54Yossi Matias:That's awesome. When I just started. I love the fact that we can now actually identify opportunities, make the scientific progress, and then apply them back and have impact. Now, the fact that we can do that at scale, at pace, and at impact that is unprecedented, I think, makes it the golden age of research. And when I think about how we can take AI and, for example, use it for accelerating scientific research, empowering healthcare workers, empowering teachers, practically empowering everyone in what we're doing, I think this opportunity is greater than ever, and we are beyond an inflection point of actually seeing this acceleration.

2:37So do you think we're going to see some... It's like we're talking a lot about it, but in terms of seeing tangible results, because we've seen what I've read about is one of the rarest diseases was cured with AI because they were able to modify the gene. Do you think we're going to see something like that in 2026?

2:56Yossi Matias:Oh, definitely. So, for example, one area that I'm really excited about is how we've built a system called AI Co-Scientist, which essentially is a way to empower researchers in their own scientific discovery. And the way it works is that a scientist can pose a question and the system, which is based on AI agents, can essentially do the work that today you expect typically, you know, research lab assistants, graduate students, postdocs to do literature research, looking into generating hypotheses, ranking them, trying to validate them and then proposing that back to the researcher. And what we found out is that this actually can play a role of an amazing collaborator.

3:40Yossi Matias:So, for example, in partnership with Imperial College, they were looking into a superbug and looking into some hypothesis of what's actually making it more infectious. And they came up with a hypothesis over many years that was not yet published. They used our system and they got a similar hypothesis in just about three days. And the system also provided another hypothesis. And when I was speaking with one of their investigators, the researchers, how does he see the experience? He said, this is an amazing collaborator for him. And similarly, we had partnerships with researchers in Stanford and other places on areas such as drug repurposing for AML or novel hypothesis for liver fibrosis.

4:24And when you think about the opportunity for having AI to be practically a virtual lab for every researcher, so you don't need to be a head of lab after 20 years of research, established research,

4:39Yossi Matias:in order to actually have your own lab. Because in the future, every junior researcher, perhaps even a grad student, can have their own virtual lab that helps them ask bigger questions, make progress faster. That's why I think about AI as an amplifier of human ingenuity. It's really a tool that is going to enable many more actually doing that. And we'll already start seeing that. There's a lot of interest in this. There are additional AI tools in advancing research. And again, this is something that is already happening. So I certainly expect to see more of that this year and certainly have impact on areas such as healthcare and certain other areas.

5:21Yossi Matias:I certainly expect that to help in material design, in finding new sources for energy in the future, and amplifying both human capacity and AI as an amplifier of human ingenuity. 100%. I'm hoping we're going to get more discoveries because of that, because we're enabling basically more people at different levels to do their own research with the help of AI. When it comes to AI being a collaborator, do you think trends like agents and vibe coding are overhyped? Because from my perspective, everyone's talking about AI agents, but making it work for you, like a non-technical person, really hard. Same with vibe coding, right?

5:59Yes, you can vibe code a simple website, but have we actually used a product that was vibe coded? I'm not sure. Refreshing Wild Cherry Cola meets Smooth Cream.

6:13The treat you deserve. Pepsi, Wild Cherry and Cream. Treat yourself.

6:21Yossi Matias:I think they're underhyped. Oh, you think so? Yeah, definitely. And here's why. So first, we should always keep in mind that what we're seeing today is not the future. We're just seeing today. I mean, many of the things we're just discussing now were not even considered just a couple of years ago. So things are making progress very fast. it's true that some of the interface still requires more work. In fact, if anything, conversational experience is the ultimate user interface. So I've been actually passionate about conversational experience for years now, developing an early AI system called Google Duplex and looking into other conversational experiences because ultimately this is the most natural user interface.

7:04Yossi Matias:In fact, we're all familiar with those who struggle to learn how to use new systems. but anybody can just say what they need and what they want and get the system to do it for them. We start seeing that with the day-to-day consumer part. And, you know, having been working for over a decade on search, in search we suddenly take the user intent and we try to give the best information. Now with AI systems we can actually understand much better user intent and let them actually express what we do. Now, Vibe Coding is a great example where people can now actually say what they want to be developed and everybody can now develop an application that previously required a team.

7:47Yossi Matias:Of course, typically it's not yet ready application to be used and there's more that needs to be done. There are actually companies out there that are providing tools to develop those applications in an easy way and we're obviously making a lot of progress on that. Even we recently shared something that we call generative UI that enables you to actually, for any prompt, get a result out of, we have an experiment within a Gemini app called Dynamic View that gives the full interaction, full interactive user interface in about a minute for any prompt that you ask, as if somebody was developing an application on your behalf.

8:27Yossi Matias:So only you need to say what you'd like to learn about, and it will sometime even give all the buttons and simulation. And in fact, the technology is now available also in search AI mode. These are, in a way, nascent. It's a nascent space. These are first technologies. But if I try to do a little fast forward and we make some of those experiences that you highlighted more intuitive, more accessible, more easy, we're actually getting to a stage that every person can actually translate their ideas, their creativity into something that is workable. I think it's extremely powerful. Obviously, for those who are going to go deep into technology, those who are going to spend time on science, those who are going to do additional degrees, there are also many opportunities to amplify.

9:19Yossi Matias:But even for those who don't, who don't necessarily have all the technical background, they are going to be able to actually build on these tools and find ways to innovate and address problems or just objectives that they have. So do you think we're entering the age where everyone is just vibe coding tools that they need every day? Or do you still feel like, because when I think about this, right, would I, if I need to jump on a call, for example, and I don't know, I have some idea in mind, it is still easier for me to use Zoom or any other tool, Google Meet, right? Because it was created by professionals who spent hours and hours and hours building it.

9:59do you think AI is going to get to that stage where a vibe-coded app will be similar to what's on the market?

10:05Yossi Matias:So one notion that I've been quite passionate about for quite some time is what I would call ambient intelligence, which is that you have technologies that you just use. You don't think about them. They're becoming so available and so intuitive that you actually don't, you just assume they work, right? Think about Autocomplete, which I also had the privilege to develop with my team over the years in search. People just assume that you start typing and it will just suggest to you. And I remember that people were in the early days, what about this magic? Wow, how does it make a guess? And today you just expect it to be the case, right?

10:44Yossi Matias:And similarly, you can think about so many other technologies. Think about voice technology. The fact that now you can speak and, you know, you expect what you're speaking to be understood. You can take text, you can listen to it. I remember actually working on these technologies in the early days. This was the aspiration. And now you just assume this is the case. Think about multiple languages. I still remember the day that one of my kids came from school and say, hey, dad, what's going on with your translate? That line, that sentence was not translated very well. It's actually quite bad translation.

11:21Yossi Matias:And I was thinking to myself, you know, just a few years ago, having automatic translation of a page was science fiction, and he assumes this is just available. Now we're much further. Now we're getting to the stage that we can actually hold a conversation. We can consume media in a different language, and it will sound in our own language, perhaps the same voice, perhaps the lips are going to sync. It's not entirely perfect yet, but if you do a little fast forward, it's going to be the technology and AI are removing all these barriers of modality languages, even interactions you know, education which is one of theirs and I think is highly important is a way that is really poised to be transformed and we have recent experiment of asking ourselves can we reimagine the textbook?

12:14Yossi Matias:Can we take a textbook and use AI in order to actually give it in different experiences that are going to be personalized and contextualized. So, for example, can I take the text and make it immersive? Think Harry Potter. Can I make it conversational? Can I have a sketchbook with that? A podcast. And by the way, can I have it in a level that is suitable for the audience? So, for example, can I explain gravity to a 10-year-old who likes soccer so that the textbook can actually be re-leveled to 10-year-old kind of language and give examples from soccer? The answer is yes, we actually have some experiment.

12:52Yossi Matias:And these are early days, so I expect that in the future it's going to be seamless. It's going to be kind of just available to us. What is the time range that you think is going to be critical when we won't be able to recognize the reality? Because from what you're saying, right, it seems like in five years, what is the timeline, do you think, for all these, for AI to be intelligent enough to pose questions for research, create those apps on the go, adjust education for anyone according to their level. So I think in a way we're leaving that, right? I mean, we're still early, right? We're still seeing the worst version of it.

13:32Yossi Matias:You know, there's the saying that advanced technology is indistinguishable from magic. But the flip side of that is that what appears to us magical now we're get used to it very quickly and we just assume it's working. So I would argue that there are many things that we're already doing that, you know, they were very novel and revolutionized perhaps at some point. And I would just assume they're working. And we're already in the midst of doing that. Think about the fact that you can actually today use generative AI in order to, you know, express some thoughts and get a summary. You know, we deep think and think of this sort.

14:11Yossi Matias:Think about the fact that you can actually take text and just listen to that. I do that a lot, by the way. I love this multimodal approach for consuming information. And so in a way, we're in the midst of that. And this keeps advancing all the time. And what do you think is going to happen to our lifestyle? So, for example, let's take, I don't know, social media marketer today, nine to five doing the tasks. Of course, a lot of them are being automated. it um ai can now write text create scripts uh what do you think their life is going to look like in five years do you think we're going to slow down like do more hobbies or because we have more tools because this is what i feel when i see how i can 10x myself with ai i'm now on 200x myself because it never stops the ambition right yeah so in a way i remember i was asked so how do you think uh things are going to change five years and and i thought about it in one hand everything is going to be different.

15:04Yossi Matias:On the other hand, nothing is going to be different. I've heard that at Davos this year. In a way, I mean, we're humans, right? And what's motivation for what we're doing? It's not really to fulfill a particular predetermined task. We define our own ambitions, our own tasks. Same with learning. I remember the early days when suddenly Google was made possible for everybody to get facts and people say, wait a minute, what's going to do for kids because we ask them to do homework and collect facts in library and now it's easy are they going to be lazy well no because now this is a given that's a tool so now we expect them actually to go to the next level we expect them to synthesize and now with AI of course there's another conversation what is it going to do and my prediction is that in fact we're just going to up level what we expect that's why I'm thinking about AI as an amplifier for human ingenuity Now, what are we going to do with that?

16:01Yossi Matias:That goes back to the motivation. Why are we doing what we're doing? Probably most of us are spending more time in our work than we should to fulfill the tasks just because we're ambitious about what we are excited about, what we can achieve. These motivations are not going away. So in a way, are lifestyle going to change? Definitely. But the basics are probably not going to change at all, which is about people working with other people, about solving important problems that they're excited about. The trend that I'm seeing now as a mom, though, is that now kids are expected to know how to read when they start school, right?

16:38Because back in the day, you're like, your kid goes to school, no letters, no numbers. Now they're like, oh, I actually have the classes already reading and they're five years old. And you're like, whoa, okay.

16:48Yossi Matias:Kids are already smarter, I think, than the older generation. And I think the next generation is going to be even smarter because they're going to have AI in their disposal. Yeah. So and by the way, you know, typically we had to most people had to focus and learn certain subject and focus mostly on that one. And when they wanted to work across disciplines, they had to meet with other folks and try to somehow do that together. We're going to have everybody's going to have a polymath in their pocket. Now, if you want to understand how something you're doing is related in another field to chemistry, to biology, to physics, you can actually get this advice right away from the AI in the right level.

17:29Yossi Matias:In the future, it's going to be even smarter connecting it to what you want. So thinking about this cross-discipline, and I'm a great fan of cross-disciplinary connecting worlds between science and technology and the arts and philosophy and so forth. I think that we are all going to benefit from actually having kind of everything in our fingertips and also be adjusted to the level that we're familiar with that. So I think there's a great opportunity here for, if you will, not only formal learning, but learning for all of us and evolving and a richer experience across everything that we do, hopefully.

18:08Yeah. And as a person who's constantly hiring, can you talk about the skills that you're looking for, both technical and non-technical?

18:20Yossi Matias:Yeah, I think it's a great example. So a great question. So, for example, when I think about in a way, we always looked at Google for people who have very strong foundations. And when you think about what are you looking for in terms of skills and in terms of education, one thing that I always thought is critical is the ability to think, the ability to adapt, the ability to evolve, the ability to actually think about problems and then try to solve them. Now, today, these are more important than ever, of course, because, you know, technology is moving fast. People need to adjust their learning, even no matter how experienced they are.

19:01Yossi Matias:There are new technologies. Engineers need to relearn how to use AI in order to be more productive, and people are doing it. So the ability to adapt, to learn, to have a strong foundation is more important than ever. How do you learn it? Any tips? How do you learn to adapt? Yeah, I think it's something that is inevitable for all of us to learn, and certainly if there's one skill that is important for a younger generation is to learn how to learn. To some extent, this is not different than I remember years ago, there was a question of saying computer science, should one learn more programming languages or more foundations, more algorithms?

19:42Yossi Matias:I always thought that it's not necessarily the specific language that is important, but more the foundational principles of how to think about matters. Now, the good news is that we can also use technology to help learning and to help adapting. So, for example, when I refer to AI as an amplifier of human ingenuity and the opportunity for every researcher to be able to use AI in order to actually ask your questions, it's important to also adapt and learn how to do that. If we really want people to be able to do the kind of roles that today we expect many more senior people who are much more senior to do, Obviously, we need them to learn how to do that much quickly.

20:28Yossi Matias:The good news is that I believe that AI can help us with that as well, with kind of tutoring, with feedback. The fact that to write a paper, for example, AI can give you the kind of feedback that in the past you actually needed the attention of your advisor perhaps to do. Oh, yeah. So I think things are going to adapt in all of these dimensions. And I think humans are, what we proved over the years is that we're extremely adaptive to what's required to do. And whenever we're setting the right goals, the right opportunities, then we see this kind of great results coming out of that. So I think we should build into that.

21:09Yossi Matias:When I say AI is an amplifier of human ingenuity, this is not only a prediction, this is a design goal. This is how I'd like us to build our systems, how I'd like us to see how we're helping the society to actually do that, how we're influencing education so that our kids can actually grow into this future. And I'm quite optimistic about how we can actually have those next generation solve many other problems in the world. Yeah, because they have more to work with. I just wanted to ask you one last question. If somebody wants to remember one thing from this conversation, what is the mindset that they should adopt for 2026 to stay as positive as you are and also relevant in the job market.

21:54Yossi Matias:Yes, so one exciting thing is about using technology and research is that certain problems that seem impossible are not necessarily impossible. In fact, I've yet to see something that is impossible to tackle. And some of them actually are quite meaningful. Here's an example. One area that we're using AI quite a bit is on climate resilience, how to address natural disasters. You know, people are turning to Google whenever they are, if you look at Google Trends, whenever there's a natural disaster, you see a peak, people are coming to ask questions because they'd like to know what they should be doing.

22:29Yossi Matias:And we actually, you know, provided some, you know, the kind of experiences in search and maps years ago and things such as we call SOS alerts. One thing I learned at the time is that one area that were not very helpful is an area of natural flood prediction. Floods are, you know, causing thousands of deaths every year. And when I asked around, all the experts told me this is impossible to solve. There are too many variables. And we decided, well, it's important enough to try anyway. Fast forward, we now have a system that provides flood predictions in 150 countries covering 2 billion people, predictions up to seven days in advance.

23:15Yossi Matias:And from what seemed to be impossible just seven years ago, we have a system doing that. we had to actually drive the research to work with the community. We published a paper in Nature doing that. So we actually took a problem that seemed impossible and made it possible. Similarly, when we think about healthcare, they are looking into the question, can we take these language models and that obviously require more work about being more factual and more trustworthy in areas such as healthcare, and showing that we can actually bring them to the level that they can pass you a style medical exam in a passing score and later an expert level.