MC INTERVIEW Google DeepMind's Pushmeet Kohli explains why India is in perfect position to leverage AI | Read full Q&A

Pushmeet Kohli, vice president of research at Google DeepMind with Moneycontrol's Nalin Mehta

• AI speeds up science and democratizes research worldwide, including India.
• AlphaFold by Google DeepMind transformed protein research, benefiting millions globally.
• India poised to harness AI for healthcare, agriculture, and innovation

Did our AI summary help?

Pushmeet Kohli, one of the "hottest" AI researchers in the world currently and vice president of research at Google DeepMind, gave an exclusive interview to Moneycontrol's Managing Editor Dr Nalin Mehta and shared his thoughts on artificial intelligence and its impact on India.

We are delighted to be joined today on Moneycontrol by Pushmeet Kohli, one of the hottest researchers in the world on AI. He is distinguished scientist and VP of research at Google DeepMind. His teams have developed many path-breaking innovations on AI, including the Nobel Prize-winning AlphaFold model for protein structures. He's done significantly path-breaking work across biology, across materials, across fusion, math and software. I'm delighted to have you with us, Pushmeet, joining us from London for this conversation with Moneycontrol.

Kohli: Thank you for having me. It's a great pleasure. I will not describe myself as a hot office researcher, but thanks for the description.

So, Pushmeet, we are talking at a time when there's a lot of flux, the great geopolitical flux. The India-US trade deal has just been signed. There's been a lot of debates around movements, tech, around deep tech, around AI and so on and so forth. Let me begin by asking you, what is your advice to young Indian researchers? Should they come to the West to build foundational models or is it now possible to do world-class AI for science research from India?

Kohli: Yeah, Nalin, I mean, it's a time of a great transformation. We are seeing an incredible technology taking shape today in the world. AI is going to revolutionize the world. It's going to sort of let us make breakthroughs in a number of different areas. I particularly sort of focus on science and we are seeing a lot of a huge transformation happening in how science is happening in the research lab today. AI is really accelerating science, democratizing science and expanding what we are seeing around the world.

Now, coming back to your question in terms of what's the advice I would give to youngsters today, the clear advice is you have to leverage this technology. You have to understand it, you have to understand its strengths, but you also have to understand its weaknesses. And in terms of how you should look at sort of AI and where you should sort of work, I think the key idea is that you have to think about the application. How is this technology going to affect the real problems that people are facing today? And in that sense, I think India is in a very good position. It has a lot of sort of opportunities to take this technology and bring benefits to the whole country. Like if you look at healthcare, if you look at agriculture, this technology can really help India advance in a number of different areas at pace.

Right. So, Pushmeet, Google is of course, stepping up its fundamental AI research in India across a range of real world challenges in areas like agriculture, maternal health, patient care, disease treatment and so on and so forth. So, looking ahead, what role do you see India playing in Google's future research as it addresses some of the biggest challenges and within the larger AI ecosystem globally? I mean, there's constant debate about whether India should focus more on services or more foundational models. What is your thinking on those debates?

Kohli: Yeah, I think, as I mentioned, I think foundational models are a key technology, advances will happen. We have a lab in Bangalore, Google DeepMind has a lab in Bangalore, which is making fundamental sort of contributions to Google's foundational model, Gemini. And not only sort of improving the foundation model, the lab is also thinking about what are the different applications that will be affected by these models. So, in agriculture, we are really thinking about how do we organize India's agricultural data and building a foundation and unified landscape understanding that can help not only the planners, but also sort of farmers really get the best out of this data.

The other sort of opportunity for India is to leverage these foundational models to understand the different languages that India has, right, and to bring in that language into that language data into the foundational models. And that's another sort of key element of the work that our Indian lab is doing.

Where do you think AI has delivered the most unreasonable progress for want of a better word?

Kohli: So, there I'm really partial to science, like I lead the science and strategic initiatives unit at Google DeepMind. And one of the very first projects that we started almost eight years back was on using AI for this amazing problem of protein structure predictions. Now, proteins are the building blocks of life. These are the Lego blocks that make you and me. And understanding their function and shape is an incredibly important sort of problem to tackle if you want to develop new treatments for diseases and so on. And our AI model AlphaFold made a major breakthrough at the height of the pandemic, in fact, in 2020, when we released AlphaFold 2. And since then, it has been used by 3.3 million researchers across the world. People have been sort of using it from developing better vaccines to coming up with new treatments for neglected tropical diseases to coming up with enzymes that can deal with plastic pollution and making sort of breakthroughs in fundamental biology. So, you can see the real impact that this kind of technology can have over a large spectrum of applications.

So, what is the, in your view, what is the next biological frontier for AI to crack?

Kohli: I think the key sort of challenge now is if proteins were the building blocks and were the ingredients of life, the key question now is to decipher the recipe book of life, our genome. Our genome is, our DNA is basically, it captures a lot of information about life, about how we sort of develop, what diseases we are susceptible to. And scientists have, over the last many decades, have been incredibly sort of successful in building technologies that can read the genome. We completed the human genome a few sort of, almost a decade back, but we are yet to really decipher it and understand the meaning of the genome. So, that's one of the grand challenges that we are taking on.

We really want to understand what is the effect of every possible change in our genetic code and what are the implications for diseases, how can that information be used for not only diagnosing diseases but coming up with better sort of treatments.

So, Pushmeet, do you think that AI can be this great equalizer in science? In many ways, it is becoming one. But are we, in a sense, moving to a future where a researcher in a small Indian university with just a laptop and your models can compete directly with the biggest labs in the West? Is that possible or is that not really, you know, there's a big gap between the rhetoric and the reality?

Kohli: Exactly, Nalin. I mean, in fact, AI's biggest potential is in the democratization of science. If you think, I mean, go back to the example of AlphaFold, to understand the structure of one protein, it used to take almost five years and multiple millions of dollars, right? You had to take the protein, synthesize it, purify it, crystallize it, then sort of send it to a synchrotron, get a diffraction pattern, and then you will find the structure. A researcher working in India or sort of Latin America or Africa working on a tropical disease never had access to a structural understanding of the proteins that they really want to study.

And now all of that millions of dollars and multiple years has been transformed into one click. So you go to the AlphaFold database, you put in your sort of protein, and you get the structure out. And that transformation, that acceleration, that democratization is what sort of is a perfect example of what AI can bring to science and to the world in various different domains.

Right. So AlphaFold, of course, revolutionized biology by predicting protein structures. It won the Nobel Prize. Where do you see the next AlphaFold moment? Is it in clean energy? Is it in climate modeling or something else?

Kohli: Yeah, so what we are seeing now in AlphaFold was basically the great example which started now this whole sort of wave of new discovery. If you look at multiple areas of science, whether you look at meteorology, where some of our latest generations of model like WeatherNext, they are able to predict weather much more accurately and much more efficiently. What used to take sort of a whole model supercomputer for a day to compute, it can now be done much more accurately on a single sort of chip in a matter of a few minutes. So that acceleration is happening in weather. It's happening in fusion research. We are sort of moving towards more practical sort of fusion energy. That acceleration is happening in quantum computing. That acceleration is happening, of course, in biology. It's also happening in material science.

We are actively sort of working on sort of models which can come up with new materials that can tackle some of the grand challenges that we are facing in terms of dealing with better sources of energy, like better battery technologies, better sort of rare earth free magnets, a number of things that are happening that are important for the world.

Right. So see, fusion is often described as being, say, almost 30 years away. So what role can AI realistically play in bringing fusion closer and sort of collapsing that timeline?

Kohli: Yeah, so I think that has been a mean that fusion will happen in 20 years. And after 20 years, again, the same sort of thing is repeated. But I think with AI, you really can break that mean. We are seeing a lot of advances in AI, sort of AI learned sort of policies, learning to stabilize plasma much more sort of efficiently. In Tokamak, this is something that my own sort of lab has done at Google DeepMind in partnership with our partners in Lausanne, EPFL. And we're working with sort of some of the private sector as well as public sector sort of entities in the private sector, common fusion systems have been developing a number of technology and we are partnering with them, public sector institutions like ITER, and there are other sort of organizations as well, which are working in the space and AI is really helping. So I sort of see a lot of advances in the next sort of few years, and really a move towards practical fusion energy generation.

So Pushmeet, you kind of talked about this earlier in this conversation. But I want to kind of ask you this question again, that how is Google DeepMind working to ensure AI breakthroughs actually reach the last mile? I mean, bridging the sort of the gap between the top end of the research universe, as it were, and people working around the world, how do we bridge the last mile both in terms of the dissemination of the research, but actually applications to be built on top of that for real life for real world solutions?

Kohli: Yeah, so I think this is a very important sort of question, right? At the end of the day, this technology will have the impact when it gets in the hands of users. So both for sort of our specialized models as I'm sort of, I can share models like AlphaFold are being used by 3.3 million researchers, because we have been able to share the predictions. But more importantly, the more broader generation of models that Google is developing now with Gemini, that is a model which is extremely powerful, it's extremely general, and it's being used across the world. It's integrated into sort of Google search in AI overviews, it's available in the Gemini app, we are using it as to power some of our most advanced scientific agents like AlphaEvolve, co-scientists. So all these sort of things and all these products are coming into the hands of users, both consumers as well as enterprises through Google Cloud.

So, Pushmeet, is there one scientific mystery that you personally wish AI could solve in your lifetime? What would that one thing that you really wish for be?

Kohli: Yeah, I mean, that's a very tough question. It's like choosing between your children. Like I can't give you one, I can give you two. In my lifetime, I would love to sort of see AI make progress towards a room temperature superconductor. I mean, that would be an incredible piece of technology that will really transform our ability to do many different things, like store energy, come up with high sort of magnetic field resistance for fusion and many other things.

And then on the other side, I really have a soft spot for biology. If we can decipher our genome, I think it will tell us much more about who we are as a species. And I think it will really sort of elevate humanity's understanding of itself and the nature around us.

Right. And finally, Pushmeet, I mean, we are seeing huge investments by Google in India at multiple different levels. Where do you, when you take a step back, see India's evolving role in the wider AI landscape? What are your views on that in terms of-- at one level, there is this whole testing on scale at population level, that is one kind of thing, then there are investments happening. But in terms of innovation, in terms of deep tech, in terms of the real next leap forward, how do you see India's place in the larger landscape?

Kohli: So, India has, is in a great position to leverage this new technology. It is a very young country in terms of the population, it has a growing sort of young population that can really learn and leverage this technology to meet its aspirations. It has a number of sort of challenges as well, right? How do we provide healthcare, high quality healthcare to a very large population? How do we democratize sort of knowledge and intelligence? This is something that India can really sort of leverage this AI movement to bring sort of high quality healthcare to the masses, to sort of the idea is basically AIIMS level sort of expertise in health available at in the rural sort of sector, sort of an amazing sort of amount of technology to deal with challenges in agriculture, in energy, and so on.

So, I think India is in the perfect position to really leverage this technology. I think the key element is education, awareness, making people aware of what the technology is, what are its strengths, but also what are its weaknesses, because only with a fuller understanding of the technology can you responsibly use it for the greatest benefits.

Right, Pushmeet Kohli, thank you so much for making the time and for speaking to us at Moneycontrol and we look forward to catching up with you at the Global AI Summit as well in Delhi. Thank you so much.

Kohli: Thank you, Nalin.