#51 - Can AI create a golden age of scientific discovery? - Yossi Matias

In this episode of The Deep View Conversations, we sit down with Yossi Matias, head of Google Research, to explore how AI is transforming the way science gets done.

Rather than replacing scientists, Google is building AI systems designed to amplify human ingenuity. From searching millions of research papers to generating new hypotheses and accelerating experiments, these tools aim to help researchers move from ideas to discoveries faster than ever before.

Yossi explains why he believes we're entering a new era where AI can democratize scientific research, empower the next generation of scientists, and dramatically shorten the path from breakthrough to real-world impact.

Topics covered include:

• How Gemini for Science is changing research workflows
• What AI Co-Scientist, AlphaEvolve, and the Empirical Research Assistant actually do
• Why the scientific method is becoming even more important in the AI era
• How Google is partnering with universities including Stanford and Imperial College
• Why AI could give every researcher a "virtual lab" in their pocket
• What a golden age of scientific discovery might look like

If you're interested in AI, scientific discovery, biotechnology, or the future of innovation, this conversation offers a look at how one of the world's leading AI research organizations sees the next decade unfolding.

Subscribe to Deep View Conversations for interviews with the leaders shaping the future of AI, business, and technology: tdv.transitor.fm 

And don't forget to sign up for The Deep View daily newsletter. We don’t just cover AI, we decode it. In a world flooded with hype, we deliver sharp, no-nonsense insights to keep you ahead of the curve and help you put AI to work every day: subscribe.thedeepview.com

Creators and Guests

Host

Jason Hiner

Editor-in-Chief of The Deep View

What is The Deep View: Conversations?

From frontier labs and enterprise platforms to emerging startups reshaping entire industries, The Deep View: Conversations podcast interviews the brightest minds and the most influential leaders in AI.

Yossi Matias: When I think about the research process, the scientific method is more important than ever. Really a lot of that is building the foundation so that others can actually build on that, both the research and applications.

One thing I'm really excited about is what I like to call the magic cycle of research, which essentially is identifying those research questions that could make a difference in products, in business, in science, in societal impact. Have a breakthrough research and then turn it into reality, which just generates an expression.

In a way, if you think about science, science is a lot about this kind of magic cycle. The exciting aspect is that magic cycle today is faster than ever and in a way we're accelerating it with all these tools. So in a way, this makes it what I think about it is the golden age of research, because the impact that we can have and the ability for us to have impact on everything from products, experiences, science and societal impact is greater than ever.

Jason Hiner: In this episode, I talked to Yossi Matias, head of Google Research, about one of the biggest ideas announced at Google I/O this year: using AI to accelerate scientific discovery.

Yossi explains how Google is building tools like Gemini for Science, AI Co-Scientist, AlphaEvolve and the Empirical Research Assistant to help researchers search literature, generate hypotheses, validate ideas and build models much faster than before.

He makes the case that AI can become an amplifier of human ingenuity, giving students, scientists and research teams the equivalent of a virtual lab in their pockets.

We also talked about why the scientific method matters more than ever in the AI era, how Google is working with institutions like Stanford and Imperial College, and why Yossi believes we may be entering a golden age of research, where the cycle from question to breakthrough to real-world impact moves faster than ever.

So here it is, our conversation with Yossi Matias of Google Research.

Yossi, thank you for coming on The Deep View Conversations podcast. You all announced a lot, and your team at research here at Google has a lot that was involved with the things that were announced here at Google I/O today. So talk a little bit about the most important things.

Yossi Matias: Well, thanks for having me here, and an exciting moment indeed.

Google Research, we're actually working across many products and many initiatives, so a lot of what was discussed is actually something that makes the team proud. As an example, some work on generative UI from last year actually made it into, with announcements, into products.

Perhaps I'll focus on one thing that's really exciting, which is how we're using AI to accelerate scientific discovery. And what we shared today is what we call Gemini for Science, which has essentially a few experimental tools that are hopefully going to accelerate scientific discovery itself, including hypothesis generation, which is built with AI Co-Scientist, and the model discovery, which is built with AlphaEvolve and Empirical Research Assistant. And we have also literature insights, which is built on NotebookLM.

And if I can double-click on at least a couple, AI Co-Scientist is something we actually announced last year. It's research which is looking into a multi-agent system that is helping a research scientist in the research by allowing them to actually do the literature search, by creating a hypothesis, by validating the hypothesis to the extent possible, and essentially squeezing work that could be done over weeks, months or years into days.

And also really excited about the fact that today we had the research paper on Co-Scientist published in Nature.

And another big, important research, I think, that leads into what we call model discovery is Empirical Research Assistant, ERA. This one is about how to help out with some of the most heavy lifting tasks that we have when we're actually doing science exploration, which is building models.

And the way ERA works is that for any scoreable problem, problem for which we know what we're aiming to actually compute, and any given input for that, it helps sift through possible models, searching through them, trying them out empirically, and coming up with proposed models that could actually accelerate the science.

Now, today we also have a Nature paper of the research on ERA. So we have these two research efforts. One is Co-Scientist, one is ERA, both of them published in Nature. And they're actually the basis for these experimental tools that are helping out, on the one hand, create hypotheses, do a literature search, by the way, cross-disciplines. So it's like having a polymath in your pocket and validating some of these hypotheses.

And on the other hand, we have also the model discovery, which allows actually to accelerate building models. And there are already quite a few papers actually built on both systems, about eight papers or so just on the ERA, in areas, disciplines, anything from epidemiology to engineering, to economy, to physics, to cosmology, etc.

So these are part of our efforts to see how to actually accelerate scientific discovery. And there's more work that we're doing in this space, for example, how to help with reading papers and so forth.

Jason Hiner: Very good. So when you think about the ways that the research team is using AI and using agents to explore their work, am I understanding it correctly that it enables them to save time or to not go down paths that they would have otherwise, it would have taken a while to decide, "Oh, this isn't the right path." But you can use AI to help model out some of those things. And so then you can choose the right path without having to do work that maybe was lost or didn't have as much value. Am I understanding that correctly?

Yossi Matias: It's true, but I think it's even more foundational than that.

Jason Hiner: Okay.

Yossi Matias: That's the scale and the depths actually make it to a new level, bring it to a new phase.

Jason Hiner: Okay.

Yossi Matias: The way I think about it is that first I think about AI as an amplifier of human ingenuity, because really what it does is practically providing everyone, and everyone, I mean junior scientists, postdocs, even grad students, they can all have their own virtual lab.

Jason Hiner: Okay.

Yossi Matias: When you think about what we're doing with these systems, they're actually doing tasks that today actually sometimes take much of the time of, say, graduate students or junior researchers.

The fact that you can actually do literature search for something that would take months or years because of the vast explosion of knowledge, you can do that in just hours. The fact that you can generate tens or hundreds of hypotheses, some of which would be reached by practically some of them would never be reached.

As an example, we had one of our early partners is Imperial College where researchers were looking into some bacteria, and they generated over years hypothesis that was never published. And with a Co-Scientist, that could actually get to the same hypothesis in just few days.

Jason Hiner: Wow.

Yossi Matias: And also they learned another hypothesis that was valued. It was about to explore. And when I spoke with our friends there, so how do you, what's your experience working with that? This is like an amazing collaborator.

So think about the future in which everyone has these virtual labs at their disposal. This is a huge amplifier of human ingenuity.

Jason Hiner: Wow. So there's so much more I would love to ask you about. But in terms of the things that your partners, because this isn't just for Google. You're building these tools and you want other scientists, you want to help accelerate the work of other scientists. So you have partners, particularly, that you work with?

Yossi Matias: Definitely. We're working with a hundred institutions, including Stanford, including Trig, including Berry College and others.

Jason Hiner: Okay.

Yossi Matias: Even on, when I think about the research process and the scientific method, the scientific method is more important than ever. Really a lot of that is building the foundation so that others can actually build on that, both the research and applications.

One thing I'm really excited about is what I like to call the magic cycle of research, which essentially is identifying those research questions that could make a difference. They could make a difference in products, in business, in science, in societal impact. Have a breakthrough research and then turn it into reality, which just generates an expression.

In a way, if you think about science, science is a lot about this kind of magic cycle. The exciting aspect is that magic cycle today is faster than ever. And in a way, we're accelerating it with all these tools and other forms.

So in a way, this makes it what I think about it is the golden age of research, because the impact that we can have and the ability for us to have impact on everything from products, experiences, science and societal impact is greater than ever.

Jason Hiner: Very good. With that in mind, I can't wait to come back next year and hear what are the advances that this has taken us forward and learn more. I can't wait to hear. Yossi, thank you so much for your time.

Yossi Matias: Thank you very much for having me here and really excited about it.

Jason Hiner: And enjoy the rest of Google I/O.

Yossi Matias: Very good. Thank you.

Jason Hiner: Thank you.