Sequoia Interview with Hassabis: Information is the essence of the universe, and AI will open up a whole new branch of science

Core Viewpoint

Summary: Demis Hassabis's interview at Sequoia Capital AI Ascent 2026

The connection between AI and games: Games are an excellent testing ground for artificial intelligence. By making AI the core gameplay, not only can algorithm concepts be effectively validated, but it can also provide early computational support for technology development.
The "timing theory" of entrepreneurship: Entrepreneurship should be "five years ahead of the times, not fifty." One must keenly capture the balance between technological breakthroughs and practical landing needs; being too early often leads to failure.
The evolution path of AGI: DeepMind's mission is clear and steadfast—first, to build general artificial intelligence (AGI); second, to use AGI to solve all complex problems, including those in science and medicine.
The core value of "AI for Science": AI is the perfect language for describing biology and complex natural systems. With AI simulations, the drug development cycle is expected to drop from several years to just weeks, potentially achieving personalized medicine.
The birth of new scientific disciplines: The complexity of AI systems will give rise to entirely new engineering sciences such as "mechanistic interpretability." At the same time, AI-driven simulation technologies will enable controlled experiments on complex social systems like economics, opening up new branches of science.
Information as the essence of the universe: Matter, energy, and information can be transformed into one another. The essence of the universe may be a grand information processing system, which gives AI profound significance in understanding the underlying laws of the universe.
The computational boundaries of the Turing machine: Modern AI systems like neural networks have proven that classical Turing machines are sufficient to simulate problems once thought solvable only by quantum computing (such as protein folding). The human brain is likely a highly approximate Turing machine.
Philosophical reflections on consciousness: Consciousness may consist of components such as self-awareness and temporal continuity. On the journey toward AGI, we should first view it as a powerful tool and use this tool to explore the grand philosophical question of "consciousness."

Content Overview

Demis Hassabis, co-founder and CEO of Google DeepMind, and winner of the 2024 Nobel Prize in Chemistry for AlphaFold, engaged in a broad and deep dialogue with Konstantine Buhler, a partner at Sequoia Capital, at the AI Ascent 2026 summit, discussing the path to AGI and the future landscape beyond AGI.

In the conversation, he explained why he firmly believes AGI is achievable by 2030, why the lengthy cycle of new drug development may collapse from ten years to just days, and why we should view "information" rather than matter or energy as the most core and fundamental essence of the universe. Additionally, he discussed how Einstein would evaluate the limitations of today's AI models if he were still alive, and why the next year or two will be a critical juncture for humanity's fate.

Full Interview

Host: Demis, thank you very much for coming.

Demis Hassabis: I'm glad to be here. Thank you all for coming; it's great to communicate with everyone here.

Host: It's a great honor to invite you to our chocolate factory.

Demis Hassabis: I just heard about this. I look forward to tasting the chocolate later.

Host: Awesome. Demis, let's get straight to the point. Today we have a true industry OG with us: a combination of original thinker, founder, and visionary, a pioneer in all fields of AI. Demis is a pure believer and a pure scientist.

Demis's Original Intention and Inner Line

Our conversation today will start with the early story of DeepMind's founding, then delve into science and technology, and finally move to audience questions. So let's get started.

Demis, you were a chess prodigy, a founder of a gaming company, and a neuroscientist. You are the founder of DeepMind and now lead a large and influential company. These identities seem unrelated, but you have mentioned that there is an inner line that runs through them all. Can you share that with us?

Demis Hassabis: There is indeed a main line, though this may have a bit of post hoc reasoning. I have long been eager to dive into the field of AI. I recognized early on that this is the most important and interesting career I could pursue in my lifetime. Since I was 15 or 16, every learning direction I chose and everything I did was aimed at one day establishing a company like DeepMind.

Games: The Training Ground for Artificial Intelligence

I entered the gaming industry through a "curve-saving" approach because, in the 1990s, the cutting-edge technology was being nurtured there. Not only AI but also graphics rendering and hardware technology. The GPUs we use today were originally designed for graphics engines, and I was already using the earliest GPUs in the late 90s. All the games I developed, whether for Bullfrog or my own company Elixir Studios, made AI a core gameplay mechanism.

My most well-known work is "Theme Park," which I developed around the age of 17. It's a theme park simulation game where thousands of little people flood into the park, enjoying various attractions and deciding what to buy in the shops. Beneath the surface of the game runs a complete economic AI model. Like "SimCity," it was a pioneering title in its genre. When I saw it sell over 10 million copies and witnessed how much joy players derived from interacting with the AI, it further solidified my determination to devote my life to the field of AI.

Later, I turned to neuroscience, hoping to draw inspiration from the workings of the brain to derive different algorithmic ideas. When the best time to found DeepMind finally arrived, integrating all these accumulations felt like a natural progression. Naturally, we later also used games as an early training ground to validate AI concepts.

Entrepreneurial Experience at Elixir Studios

Host: The room is filled with entrepreneurs today, and you must resonate deeply with them, as you have not only founded one company but have been through two entrepreneurial journeys. Let's go back to your first venture, Elixir Studios. What was that experience like? Although it is not the company you are most known for, you achieved great success with it. How did you lead that company? What did that experience teach you about "how to build a company"?

Demis Hassabis: Yes, I founded Elixir Studios right after graduating from university. I was fortunate to have worked at Bullfrog Productions beforehand. Anyone familiar with games knows that it was a legendary studio in the early days of the industry, possibly the top game studio in the UK and even Europe at that time.

I wanted to do something that could push the boundaries of AI. In fact, back then, I was funding AI research through game development as a "curve-saving" approach, constantly challenging the forefront of technology and combining it with extreme creativity. I believe this philosophy still applies to the exploratory research we conduct today.

The most profound lesson I learned is: you need to be five years ahead of the times, not fifty. At Elixir Studios, we attempted to develop a game called "Republic," aimed at simulating a complete nation. The game's premise was that players could overthrow the dictator ruling the country in various ways, and we realistically simulated vibrant, breathing cities in the game.

You have to understand, this was the late 90s, when computers were still using Pentium processors. We had to run all the graphics rendering and AI logic for a million people on the home computers of that time. This was indeed too ambitious—almost a bit unrealistic—and it led to a series of problems.

I have always remembered this lesson: You need to be ahead of the times, but if you are fifty years ahead, you are likely to fail. Of course, if an idea is obvious to everyone, it is too late to enter the market. So, the key is to find that subtle balance.

Founding DeepMind in 2009

Host: Okay, speaking of not being too far ahead of the times, let's move to 2009, when you were convinced that general artificial intelligence (AGI) would be realized. At that time, perhaps you were only ten years ahead of the times, which is better than being fifty years ahead. Talk to the entrepreneurs here about 2009. How did you convince that initial group of top talents? You indeed recruited a high-caliber team and early members. At that time, AGI sounded like science fiction; how did you make them believe in all of this?

Demis Hassabis: At that time, we keenly captured some interesting clues. We thought we were only five years ahead, but we might have actually been ten years ahead. Deep learning had just been invented by Jeff Hinton and his academic colleagues, but almost no one realized its significance. We had a deep foundation in reinforcement learning and felt that if we combined these two technologies, we would achieve breakthrough progress. Before this, they had almost never been used together—if they had, it was limited to academic "toy problems." In the field of AI, they were completely isolated islands.

Additionally, we saw the prospects of compute; GPUs were about to shine. Of course, we now use TPUs, but at that time, the acceleration of the computing industry was going to be a huge driving force. Meanwhile, towards the end of my PhD and postdoctoral career, some of my colleagues were computational neuroscientists, and we extracted enough valuable ideas and principles from the mechanisms of the brain, including a core belief: reinforcement learning could ultimately lead to AGI through scale.

We felt we had gathered these core elements. We even felt like guardians of some earth-shattering secret because no one in academia or industry believed AI could achieve any significant breakthroughs. In fact, when we proposed to focus on developing AGI—or what was sometimes referred to as "strong AI"—many academics would roll their eyes at us. To them, it was clearly a dead end; after all, people had tried and hit a wall in the 90s.

I was a postdoc at MIT, which was a stronghold for expert systems and first-order logic language systems. Looking back, it seems incredible, but at that time, I felt that approach was too outdated. However, whether in Cambridge, UK, or at MIT, the traditional AI research strongholds were still using that old method. This further convinced me that we were on the right track. At least, if we were destined to fail, we would do so in a completely new way, rather than repeating the failures of the 90s in developing AGI. This made me feel it was worth a try; even if it was a research endeavor with an uncertain future, if we ultimately failed, at least we would fail in an original way.

DeepMind's Mission and Betting on AGI

Host: Did you encounter any general resistance to your early beliefs? Did you need to prove something to yourself or them to get the early followers on board?

Demis Hassabis: Regardless of the circumstances, I would dedicate my life to artificial intelligence. It has proven to develop far beyond our most optimistic expectations. However, this was still within the scope of our predictions in 2010—at that time, we thought it was a 20-year journey.

I believe that as a member of this field, our progress has been entirely in line with expectations, and we have clearly played our part.

To take a step back, even if things had not developed this way, AI is still a niche discipline today, and I would still stick to this path because it is the most important technology in my heart. My goal is very clear; DeepMind's initial mission statement was: First, to crack intelligence, that is, to build general artificial intelligence (AGI); second, to use it to solve all other problems. I have always believed this is the most important and fascinating technology that humanity could invent.

It is both a tool for scientific exploration and a fascinating creation in itself, and it is one of the best ways for us to understand the nature of our own minds (such as consciousness, the essence of dreams, and creativity). As a neuroscientist, I often felt a lack of an analytical tool like AI when pondering these questions. It provides a comparative mechanism that allows us to conduct in-depth studies and comparisons of two different systems, much like conducting controlled experiments.

The Culture of "AI for Science"

Host: Comparing different systems. Let's talk about "AI for Science." You have been involved in this field early on, a firm believer and a pure idealist. This is the core mission driving you. How did the model and culture you established when founding DeepMind keep it at the forefront of "AI for Science"?

Demis Hassabis: This is indeed our ultimate goal. For me personally, the fundamental driving force is to build AI to advance science, medicine, and our understanding of the world. This is how I practice my mission—by creating the ultimate tool through a "meta way," and once it matures, using it to achieve breakthroughs in science. We have already achieved milestones like AlphaFold, and I believe more will emerge in the future.

DeepMind has always prioritized this goal. In fact, we have an "AI for Science" department led by Pushmeet Kohli, which has been established for nearly a decade. We officially launched this work almost immediately after returning from the AlphaGo competition in Seoul, and it has been exactly ten years since then.

I had been lying in wait, waiting for the algorithms to become powerful enough and the concepts to become general enough. For me, conquering Go was a historic turning point; at that moment, we realized the time had come to apply these concepts to important real-world issues, starting with significant scientific challenges.

We have always believed this is the most beneficial destination for AI. What could be more beautiful than using it to cure diseases, extend human health spans, and assist in medical endeavors? Following closely are key areas like materials science, environment, and energy. I believe AI will shine brightly in these fields in the coming years.

Breakthroughs in Biology and Isomorphic Labs

Host: How has AI achieved breakthroughs in biology? You have been deeply involved in the work of Isomorphic Labs, a field you are passionate about. From the beginning, you have firmly believed in AI's potential to cure diseases. When can we expect to see a "highlight moment" in biology similar to those in language and programming?

Demis Hassabis: I believe the birth of AlphaFold has already brought us a "highlight moment" in biology. Protein folding and its three-dimensional structure have been a scientific problem for 50 years. If you want to design drugs or decipher the fundamental codes of biology, conquering this problem is crucial. Of course, this is just one part of the drug discovery process; it is extremely critical, but only one part.

Our newly spun-off company, Isomorphic Labs (which I also enjoy managing), is focused on building core technologies in biochemistry and chemistry. These technologies can automatically design compounds that fit perfectly with specific parts of proteins. Since we have mastered the shapes of proteins and their surface structures, we have effectively locked onto the targets. Next, we must create compounds that can bind strongly to those targets while ideally avoiding any off-target reactions that could cause toxic side effects.

Our ultimate dream is to transfer the exploratory process, which currently occupies 99% of the workload and time in development, entirely to computer simulations (In Silico), leaving only the final verification stage for physical wet experiments (Wet Lab). If we can achieve this—I firmly believe we will in the coming years—we can shorten the average drug discovery cycle from ten years to just months, weeks, or even days.

I believe that once we cross this critical point, conquering all diseases will become attainable. Concepts like personalized medicine (for example, drug variants tailored to individual patients) will also become a reality. I think the entire landscape of healthcare and drug development will be completely reshaped in the coming years.

New Science Born from Simulators

Host: That's fantastic. You have mentioned "AI for Science" multiple times. Do you think that at some point in the future, AI will give birth to entirely new scientific systems? Just as the Industrial Revolution spawned thermodynamics. Will our education system see the emergence of fundamentally new disciplines? If so, what might they look like?

Demis Hassabis: Regarding this, I believe the following will happen.

First, the understanding and analysis of AI systems themselves will evolve into a complete discipline—an engineering science. The creations we are building are incredibly fascinating and also extremely complex. Ultimately, their complexity will rival that of the human mind and brain. Therefore, we must study them in depth to thoroughly understand how these systems work, which is far beyond our current level of understanding. I believe a brand new field will inevitably rise; mechanistic interpretability is just the tip of the iceberg, and we have vast exploration space in analyzing these systems.

Secondly, I also believe AI itself will open up new scientific doors. What excites me the most is "AI for Simulations." I am obsessed with simulations; all the games I have written not only contain AI but are essentially simulators. I believe simulators are the ultimate path to solving problems in social sciences like economics and other humanities.

The tricky part of these disciplines is that, like biology, they are emergent systems, making it extremely difficult to conduct repeatable controlled experiments. Suppose you want to raise interest rates by 0.5%; you can only operate in the real world and then observe the consequences; you can have a set of theories, but you cannot repeat that experiment thousands of times. However, if we can accurately simulate these complex systems, then rigorous sampling deductions based on highly precise simulators may establish a new science. I believe this will empower us to make better decisions in areas currently filled with high uncertainty.

Host: To achieve these extremely precise simulations, what conditions do we need? For example, world models, what kind of scientific and engineering breakthroughs do we need to reach this point?

Demis Hassabis: I have been thinking deeply about this issue. In our work, we extensively use learning simulators. These simulators are applied in areas where we either lack sufficient understanding of the mathematical principles or where the systems are too complex. We cannot solve problems merely by writing direct simulation programs for specific situations because that approach is not precise enough and cannot cover all variables.

We have already practiced this in weather forecasting. We have the world's most accurate weather simulator, "WeatherNext," which runs much faster than the tools currently used by meteorologists. I am not sure if we can understand everything, nor am I sure if that is a good idea, but the first step is to better understand these complex systems.

Even in biology, we are studying what is called "virtual cells"—an extremely dynamic emergent system. Just as mathematics is the perfect descriptive language for physics, machine learning will become the perfect descriptive language for biology. In biology and many natural systems, there are vast amounts of weak signals, weak correlations, and massive data that far exceed the analytical capabilities of the human brain. However, within this vast data, there are indeed inherent connections, correlations, and thought-provoking causal relationships.

Machine learning is the perfect tool for describing such systems. To this day, mathematics has not been able to achieve this, either because the systems are too complex for even top mathematicians to handle, or because the expressiveness of mathematics is insufficient to understand these highly emergent dynamic systems—partly because they are extremely chaotic and stochastic in nature.

Ultimately, once you master these simulators, you may derive a new branch of science. You can attempt to extract explicit equations from these implicit or intuitive simulators. Since you can sample the simulator countless times at will, perhaps one day you will discover fundamental scientific laws like Maxwell's equations.

Maybe. I don't know if such laws exist for these emergent systems, but if they do exist, I see no reason why we cannot discover them through this method.

Host: That would be remarkable. You have mentioned a theory that the fundamental building blocks of everything in the universe may be akin to information, which belongs to a more theoretical level. How do you view this? What does it mean for traditional classical Turing machines?

Demis Hassabis: Of course, you can cite the famous E=mc² and all of Einstein's research to illustrate that energy and matter are essentially equivalent. But I actually believe that information also has a certain equivalence. You can fundamentally view the organization of matter and structure—especially systems like biology that can resist entropy—as information processing systems. Therefore, I believe these three can be transformed into one another.

However, I have a feeling that information is the most fundamental. This is in stark contrast to the views of classical physicists in the 1920s, who believed that energy and matter were primary. I actually think that viewing the universe primarily as composed of information is a better way to understand this world.

If this holds true—and I believe there is much evidence supporting this—then the significance of artificial intelligence is even deeper than we imagine. It is already significant because its core is organizing information, understanding information, and constructing informational objects.

In my view, the core of artificial intelligence is information processing. If you take information processing as the primary way to understand the world, you will find that there are profound inherent connections between these seemingly disparate fields.

Host: Do you think classical Turing machines can compute everything?

Demis Hassabis: Sometimes I think about our work and see myself as a "defender of Turing," because Alan Turing is one of my most revered scientific heroes. I believe the work he did laid the foundation not only for computers and computer science but also for artificial intelligence. The theory of the Turing machine is one of the most profound achievements in history: anything computable can be computed by a relatively simple machine. Therefore, I believe our brains are likely also a form of approximate Turing machines.

It is fascinating to think about the connection between Turing machines and quantum systems. However, what we have demonstrated through systems like AlphaGo and especially AlphaFold is that classical Turing machines, dressed in the guise of modern neural networks, can model problems that were previously thought to require quantum mechanics to solve. For example, protein folding is, in a sense, a quantum system involving extremely small particles, and people might think it is necessary to consider all quantum effects of hydrogen bonds and other complex interactions.

However, it turns out that classical systems can yield an approximately optimal solution. Therefore, we may find that many things we once thought required quantum systems to simulate or run can actually be modeled on classical systems if the methods are appropriate.

Consciousness Philosophy

Host: You have always viewed artificial intelligence as a tool, much like telescopes, microscopes, or astrolabes in past centuries. However, when you face a machine that can almost simulate everything—like you said, it can even simulate quantum systems—when will it transcend the category of a tool? Will that day really come?

Demis Hassabis: I strongly feel that in the mission and journey of building general artificial intelligence (AGI), we, as fellow travelers—including many of you here—believe the best way is to first build a tool: an extremely intelligent, practical, and precise tool, and then cross the next threshold. The significance of this alone is already profound. Of course, this tool may become increasingly autonomous and exhibit more characteristics of an agent, which is what we are currently witnessing. We are in the midst of a wave of the agent era.

However, there are further questions: Does it possess agency? Does it have consciousness? These are questions we will have to face. But I suggest we treat this as a second step; perhaps we can use the tool built in the first step to help us explore these profound questions.

Ideally, through this process, we can also better understand our own brains and thoughts and define concepts like "consciousness" more precisely than we do today.

Host: Do you have any rough predictions for the future definition of consciousness?

Demis Hassabis: No, other than what has been discussed in philosophy for thousands of years, I don't have much to add. But it is clear to me that certain components are obviously necessary. They may be necessary but not sufficient conditions. Concepts like self-awareness, the notion of self versus others, and some form of temporal continuity are clearly essential for any entity that appears to be conscious.

However, what the complete definition is remains an open question. I have discussed this with many great philosophers. A few years ago, I had an in-depth conversation with the late Daniel Dennett on this topic. One of the core questions is about the behavior of the system: does it behave like a conscious system? You could argue that as some AI systems get closer to AGI, they may eventually achieve this.

But the follow-up question is: why do we believe each other to be conscious? One reason is our behavior; we behave like conscious beings. But another factor is that we all operate on the same underlying substrate.

Therefore, I believe if both points hold, then assuming your experience and mine are the same is logically the most parsimonious. This is why we usually do not argue about whether the other is conscious. But it is evident that we will never achieve the same substrate equivalence in artificial systems. So I think it is very difficult to completely eliminate this gap. You can examine it behaviorally, but what about experientially? After achieving AGI, there may be some ways to address this issue, but that may be beyond today's discussion, even in the context of "AI and Science."

Host: Wonderful. We will soon open the floor for audience questions, so please prepare your questions. You just mentioned philosophers, especially Kant and Spinoza, saying they are your two favorite philosophers. Kant is a typical deontological philosopher, emphasizing the concept of responsibility; while Spinoza holds a nearly deterministic view of the universe. How do you connect these two vastly different ideas? What is your fundamental understanding of how the world operates?

Demis Hassabis: I like these two philosophers and am deeply impressed by them because Kant proposed a viewpoint—I experienced this deeply while pursuing my PhD in neuroscience—that "the mind creates reality," which I believe is fundamentally correct. This provides us with another excellent reason to study the mechanisms of the mind and brain. Since I am ultimately exploring the nature of reality, we must first understand how the mind interprets reality. This is the insight I gained from Kant.

As for Spinoza, it is more about the spiritual dimension. If you try to use science as a tool to understand the universe, you are already beginning to touch on the deep mysteries behind how the universe operates.

This is precisely my insight into our current endeavor. When I engage in scientific research, delve into artificial intelligence, and build these tools, I feel as if we are, in some way, reading the language of the universe.

Host: Beautiful. This is the most beautiful interpretation of your daily work: Demis, you embody the roles of scientist, speaker, and philosopher. Before we conclude, let's do a quick-fire round. He has never seen these questions before. Predict the year when general artificial intelligence (AGI) will be realized: will it be earlier or later than expected? Or you can refuse to answer this question.

Demis Hassabis: I choose 2030. I have been very firm on this prediction.

Host: Okay, 2030. So, when we achieve general artificial intelligence (AGI), what books, poetry, or papers do you recommend as must-reads?

Demis Hassabis: For the world after achieving general artificial intelligence (AGI), my favorite book is David Deutsch's "The Fabric of Reality." I believe the ideas in this book are still applicable. I hope to use general artificial intelligence (AGI) to answer the profound questions raised in that book, which will also be the focus of my subsequent work in the AGI era.

Host: Fantastic. What has been your proudest moment at DeepMind so far?

Demis Hassabis: We have been fortunate to experience many peak moments. I think the proudest should be the birth of AlphaFold.

Host: Lastly, a few questions about games. If you were participating in a high-stakes turn-based strategy game, like "Civilization" or "Polytopia," and could choose a scientist from history as a teammate, such as Einstein, Turing, or Newton, who would you choose to join your team?

Demis Hassabis: I think I would choose von Neumann. After all, in such a situation, you need a game theory expert, and I believe he is the best.

Host: That would definitely be a legendary teammate. Demis, you are truly a polymath. Thank you very much for being on our show today. Please join me in applauding Demis for his wonderful insights. Thank you very much.

Join ChainCatcher Official

Telegram Feed: @chaincatcher

X (Twitter): @ChainCatcher_