Original text compiled: Brother Gua AI New Knowledge
This article's content is compiled from the interview with Demis Hassabis on Sequoia Capital's channel, publicly released on April 29, 2026.
Content Overview: Demis Hassabis Interview at Sequoia Capital AI Ascent 2026
-
AI and Games Genesis: Games are an excellent proving ground for artificial intelligence. By making AI the core gameplay mechanic, it can effectively validate algorithmic ideas and also provide early-stage computational support for technological development. -
Entrepreneurial "Timing Theory": Entrepreneurship should be "five years ahead of its time, not fifty." One must keenly grasp the balance point between technological breakthroughs and practical application needs; being too far ahead often leads to failure. -
AGI Evolution Path: DeepMind's mission is clear and steadfast—first, build Artificial General Intelligence (AGI); second, use AGI to solve all complex problems, including those in science and medicine. -
Core Value of "AI for Science": AI is the perfect language for describing biology and complex natural systems. With AI simulation, the drug discovery cycle is expected to shrink from years to weeks, even enabling truly personalized medicine. -
Birth of New Scientific Disciplines: The complexity of AI systems themselves will give rise to new engineering sciences like "mechanistic interpretability." Simultaneously, AI-driven simulation technology will enable humans to conduct controlled experiments on complex social systems like economics, opening up entirely new scientific branches. -
Information as the Essence of the Universe: Matter, energy, and information are interchangeable. The essence of the universe might be a grand information processing system, giving AI profound significance in understanding the universe's fundamental operating principles. -
Computational Limits of Turing Machines: Modern AI systems like neural networks have proven that classical Turing machines are sufficient to simulate problems once thought solvable only by quantum computing (like protein folding).The human brain is likely some form of highly approximate Turing machine. -
Philosophical Reflections on Consciousness: Consciousness might be composed of components like self-awareness and temporal continuity. On the journey towards AGI, we should first view it as a powerful tool, and then explore the grand philosophical question of "consciousness" with its assistance.
Content Introduction
Demis Hassabis, Google DeepMind co-founder and CEO, and winner of the 2024 Nobel Prize in Chemistry for AlphaFold, held a wide-ranging and profound conversation with Sequoia Capital partner Konstantine Buhler at the AI Ascent 2026 summit, discussing the path to AGI and the future beyond.
In the dialogue, he explained why he firmly believes AGI could be achieved by 2030, why the lengthy cycle of new drug discovery might collapse from a decade to just a few days, and why we should regard "information," rather than matter or energy, as the most core and fundamental essence of the universe. Additionally, he pondered what Einstein might say about the limitations of today's AI models if he were still alive, and why the next year or two will become a critical juncture in determining humanity's destiny.
Full Interview Transcript
Host: Demis, thank you so much for coming.
Demis Hassabis: Pleasure to be here. Thank you all for coming, it's fantastic to be here chatting with you all.
Host: It's an absolute honor to have you in our chocolate factory.
Demis Hassabis: I just heard about that. Looking forward to trying some chocolate later.
Host: Wonderful. Demis, let's dive right in. Today we have a true OG: an original thinker, founder, visionary, pioneer in all things AI. Demis is a pure believer, a pure scientist.
Demis's Origin and Inner Thread
Our conversation today will start with the early story of DeepMind's founding, then delve into science and technology, and finish with audience questions. Let's begin.
Demis, you were a chess prodigy, a game company founder, and a neuroscientist. You are the founder of DeepMind, and now lead a large, pivotal company. These identities may seem disparate, but you've said there's an inner thread connecting them. Can you share that with us?
Demis Hassabis: There is indeed a thread, although perhaps with a bit of post hoc reasoning. But my desire to work in AI goes way back. I decided very early that this was the most important and interesting thing I could spend my life on.From around 15, 16 years old, every subject I chose to study, everything I did, was with the eventual aim of one day building a company like DeepMind.
Games: The Proving Ground for AI
I "detoured" into the games industry because in the 90s, the cutting-edge technology was all there. Not just AI, but graphics rendering and hardware technology. The GPUs we all use today were originally designed for graphics engines, and I was using the earliest GPUs in the late 90s. All the games I worked on, whether for Bullfrog or my own company Elixir Studios, had AI as a core gameplay mechanic.
My most famous work was probably "Theme Park," developed when I was about 17. It's an amusement park simulation where thousands of little people pour into the park, ride rides, and decide what to buy in shops. Underneath, it runs a complete economic AI model. Like SimCity, it was a groundbreaking game in its genre. Seeing it sell over 10 million copies and witnessing firsthand how much players enjoyed interacting with the AI only reinforced my decision to dedicate my life to AI.
Later, I switched to neuroscience, hoping to draw inspiration from how the brain works to derive different algorithmic ideas. When the perfect moment finally arrived to found DeepMind, synthesizing all these accumulated experiences felt natural. And indeed, we later used games as an early proving ground for AI ideas.
Entrepreneurial Experience at Elixir Studios
Host: The room is full of entrepreneurs today, you must relate, as you've not only founded one company but have been through this twice. Let's go back to your first venture, Elixir Studios. What was that experience like? It may not be your most famous company, but 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: Well, I founded Elixir Studios right after university. I was fortunate to have previously worked at Bullfrog Productions. Those familiar with gaming know it was an early legendary studio, probably the best in the UK, maybe Europe, at the time.
I wanted to push the boundaries of what could be done with AI. Actually, in those days, I used game development as a "detour" to fund AI research, constantly challenging the technological frontier and combining it with extreme creativity. I think that ethos still applies to the blue-sky research we do today.
Perhaps the most profound lesson I learned is: you want to be five years ahead of your time, not fifty. At Elixir, we tried to develop a game called "Republic" that aimed to simulate an entire nation. The premise was that players could overthrow the dictator ruling the country in various ways, and we simulated living, breathing cities.
This was the late 90s, PCs had Pentium processors. We had to run all the graphics rendering and AI logic for a million people on home computers of that era. It was too ambitious—over-ambitious even—and caused a cascade of issues.
I learned that lesson well:You want to be ahead, but if you're fifty years ahead, you'll probably fail. Of course, it's too late when an idea becomes obvious to everyone. So, it's about finding that sweet spot.
Founding DeepMind in 2009
Host: Okay, on the topic of not being too far ahead, fast forward to 2009. You were convinced AGI would happen. That time, perhaps only ten years ahead, better than fifty. Talk to our entrepreneurs here about 2009. How did you convince those first brilliant minds? Because you did recruit an incredibly high-caliber group of early team members. At the time, AGI sounded like pure science fiction. How did you get them to believe?
Demis Hassabis: We had picked up on some interesting threads at the time. We thought we were maybe five years ahead, but it turned out to be more like ten. Deep Learning had just been invented by Jeff Hinton and his academic colleagues, but hardly anyone realized its significance. And we had a strong background in Reinforcement Learning. We felt combining these two would lead to breakthroughs. They had rarely been combined before—if at all, only on academic "toy problems." In the AI field, they were completely separate islands.
Additionally, we saw the promise of Compute; GPUs were about to take off. Today we use TPUs, but back then, the acceleration computing industry would be a huge driver. Also, towards the end of my PhD and postdoc, as I gathered some folks who were computational neuroscientists, we extracted enough valuable ideas and principles from brain mechanisms, including a core belief: that reinforcement learning, scaled up, could ultimately lead to AGI.
We felt we had the key ingredients.We even felt like keepers of a secret because, in academia or industry, no one believed AI would make any significant breakthroughs. In fact, when we proposed aiming for AGI—or sometimes called "Strong AI" back then—many academics would literally roll their eyes. To them, it was a dead end; people had tried and failed in the 90s.
I was at MIT for my postdoc, a stronghold for Expert Systems and First-order Logic Language Systems. Looking back, it's incredible, but even then, I felt that approach was too rigid and old. But in traditional AI hubs like Cambridge, UK, or MIT, people were still using the old methods. That actually made me more confident we were on the right track.At least, if we were going to fail, we'd fail in a new way, not repeat the 90s AGI failures. That made it feel worth trying; even as a risky research endeavor, if we failed, at least we'd fail originally.
DeepMind's Mission and Betting on AGI
Host: Did your early beliefs face widespread skepticism? What did you need to prove to yourself or others to get those early followers to join?
Demis Hassabis: Regardless of circumstances, I would have dedicated my life to AI. It has exceeded even our most optimistic expectations. But it was within our 2010 prediction—we thought it would be a 20-year journey.
I think our pace, as part of the field, is exactly on track, and we've clearly played our part.
Stepping back, even if things hadn't developed this way, even if AI remained a niche subject today, I'd still be on this path because it's the most important technology ever in my view. My goal was clear, DeepMind's original mission statement was: First, solve intelligence, i.e., build AGI; second, use it to solve everything else. I've always believed this is the most important and fascinating technology humanity could invent.
It's a tool for scientific exploration, a fascinating creation in itself, and one of the best ways to understand our own minds—consciousness, dreams, creativity. As a neuroscientist, I used to think about these questions and felt we lacked an analytical tool like AI. It provides a comparative mechanism, allowing us to study and compare two different systems, almost like a controlled experiment.
Culture of "AI for Science"
Host: Comparing different systems. Let's talk about "AI for Science." You were early, a firm believer, a pure idealist. This is a core driving mission. How did the model and culture you established when founding DeepMind keep it at the forefront of "AI for Science"?
Demis Hassabis: That's the ultimate goal. For me personally, the fundamental driver is to build AI to advance science, medicine, and our understanding of the world. That's how I execute the mission—through a "meta way": first build the ultimate tool, then use it, once mature, to achieve scientific breakthroughs. We've had successes like AlphaFold, and I believe there will be many more.
DeepMind has always prioritized this goal. In fact, we have an "AI for Science" division led by Pushmeet Kohli, nearly a decade old now. We formally started this work almost right after returning from the AlphaGo match in Seoul, exactly ten years ago.
I had been waiting for the algorithms to become powerful enough, the ideas general enough. For me, conquering Go was a historic turning point; we realized then that the time had come to apply these ideas to real-world important problems, starting with these grand scientific challenges.
We always believed this was AI's most beneficial destination. What could be better than curing diseases, extending healthy human lifespan, and aiding medicine? Followed closely by material science, environment, energy—key areas. I believe AI will shine brightly in these fields in the coming years.
Biology Breakthroughs and Isomorphic Labs
Host: How is AI achieving breakthroughs in biology? You're deeply involved with Isomorphic Labs, an area you're passionate about. From the start, you've been a firm believer in AI's potential to cure disease. In biology, when will we have our "breakout moment" like in language and programming?
Demis Hassabis: I think we already had our "breakout moment" for biology with AlphaFold. Protein folding and its 3D structure was a 50-year scientific challenge. Solving it is crucial for designing drugs or deciphering biology's fundamental code. Of course, it's just one part of drug discovery—a critical one, but still one part.
Our newly spun-out company, Isomorphic Labs (I'm also enjoying running it), is dedicated to building the core technologies in biochemistry and chemistry that can automatically design compounds that perfectly bind to specific sites on proteins. Now that we know the protein's shape and surface structure, we have the target. Next, we must create compounds that strongly bind to that target, ideally avoiding any off-target effects that could cause toxicity.
Our ultimate dream is to move 99% of the discovery process—which currently takes up the bulk of time and effort—into in silico simulation, leaving only the final validation for wet labs. If we can achieve that—and I firmly believe we will in the coming years—we can shrink the average 10-year drug discovery cycle to months, weeks, eventually even days.
I believe that once we cross that threshold, tackling all diseases becomes achievable. Concepts like personalized medicine (e.g., drug variants tailored to individual patients) will become reality. I think the entire landscape of medicine and drug discovery will be completely reshaped in the coming years.
New Science Born from Simulators
Host: Fascinating. You've mentioned "AI for Science" multiple times. Do you think at some point in the future, AI will give birth to entirely new scientific systems? Like how the Industrial Revolution gave rise to thermodynamics. Will there be essentially new subjects in our education system? If so, what would they look like?
Demis Hassabis: On that point, I think a few things will happen.
First, the understanding and dissection of AI systems themselves will evolve into a full discipline—an engineering science. These creations we are building are incredibly fascinating and also extremely complex. Eventually, their complexity will rival the human mind and brain. So, we must study them deeply to fully understand how they work, far beyond our current understanding. I believe a whole new field will arise; mechanistic interpretability is just the tip of the iceberg; there's vast space to explore in parsing these systems.
Second, I also believe AI itself will open doors to new sciences. What excites me most is "AI for Simulations." I'm fascinated by simulation; all the games I've written not only contained AI but were essentially simulators. I think simulators are the ultimate path to cracking problems in social sciences like economics and other humanities.
The difficulty with these disciplines is that, like biology, they are emergent systems, incredibly hard to run repeatable controlled experiments on. Say you want to raise interest rates by 0.5%, you have to do it in the real world and see the consequences; you can have theories, but you can't repeat the experiment thousands of times. However, if we could simulate these complex systems accurately, then rigorous sampling based on highly accurate simulators could perhaps establish a new science. I believe this would empower us to make better decisions in areas currently fraught with high uncertainty.
Host: To achieve these extremely accurate simulations, what conditions do we need? For example, world models—what scientific and engineering breakthroughs do we need to reach that point?
Demis Hassabis: I've been thinking deeply about this. In our work, we use learning simulators heavily. These simulators are applied in areas where we either don't understand the math well enough, or the system is too complex. We can't solve the problem just by writing direct simulation code for the specific case because that's not precise enough and can't capture all variables.
We already practice this with weather forecasting. We have the world's most accurate weather simulator, "WeatherNext," which runs much faster than tools meteorologists currently use. I'm not sure we can know everything, nor if that's a good idea, but the first step is to better understand these complex systems.
Even in biology, we're working on so-called "virtual cells"—an extremely dynamic emergent system.Just as mathematics is the perfect descriptive language for physics, machine learning will be the perfect descriptive language for biology. In biology and many natural systems, there are vast amounts of weak signals, weak correlations, and massive data, far beyond human brain analysis capacity. Yet, within these massive datasets, there are intrinsic connections, correlations, and thought-provoking causal relationships.
Machine learning is the perfect tool for describing such systems. Until now, mathematics couldn't do it, either because the systems are too complex even for top mathematicians, or because mathematics lacks the expressive power to understand these highly emergent dynamic systems—partly because they are extremely messy and stochastic.
Ultimately, once you master these simulators, perhaps a new branch of science can emerge. You might try to extract explicit equations from these implicit or intuitive simulators. Since you can sample the simulator arbitrarily many times, perhaps one day you could discover fundamental scientific laws like Maxwell's equations.
Maybe. I don't know if such laws exist for emergent systems, but if they do, I see no reason why we couldn't discover them using this method.
Host: That would be remarkable. You've spoken about a theory that the fundamental building block of everything in the universe might be akin to information, which is more theoretical. How do you view that? What does that imply for traditional classical Turing machines?
Demis Hassabis: Of course, you can quote the famous E=mc2 and all of Einstein's work, showing energy and matter are essentially equivalent. But I actually think information also has a kind of equivalence. You can view the organization of matter and structure—especially systems like biology that resist entropy—as essentially information processing systems. So, I think you can convert the three into each other.
However, I have a feeling information is the most fundamental. This is opposite to what classical physicists in the 1920s thought, when energy and matter were considered primary.I actually think viewing the universe as primarily made of information is a better way to understand the world.
If this holds—and I think there's a lot of evidence supporting it—then AI's significance is even deeper than we thought. It's already immensely significant because its core is about organizing information, understanding it, and constructing informational objects.
To me, AI's core is information processing. If you take information processing as the primary way to understand the world, you find deep internal connections between seemingly disparate fields.
Host: So, do you think classical Turing machines can compute everything?
Demis Hassabis: Sometimes I reflect on our work and see myself as a "defender of Turing," because Alan Turing is one of my greatest scientific heroes. I believe his work laid the foundations not only for computers and computer science but also for AI. Turing machine theory is one of the most profound results ever: anything computable can be computed by a relatively simple machine to describe. Therefore, I think our brains are likely also approximate Turing machines.
Thinking about the link between Turing machines and quantum systems is fascinating. However, what we've demonstrated with systems like AlphaGo and especially AlphaFold is that classical Turing machines, dressed in modern neural networks, can model problems previously thought to require quantum mechanics. For example, protein folding is in some sense a quantum system involving very small particles; one might think you have to consider all quantum effects of hydrogen bonds and other complex interactions.
Yet it turns out, with a classical system, you can get an approximately optimal solution. So, we may find that many things we thought needed quantum systems to simulate or run can actually be modeled on classical systems, if we go about it the right way.
Consciousness Philosophy
Host: You've always viewed AI as a tool, like the telescope, microscope, or astrolabe over past centuries. But when you face a machine that can simulate almost everything—as you said, even quantum systems—when does it transcend being just a tool? Will that day truly come?
Demis Hassabis: I very strongly feel that in the mission and journey to build AGI, we—including many here—think the best way is to first build a tool: an incredibly intelligent, practical, and precise tool, then cross the next threshold. That itself is profound enough. Of course, this tool may become increasingly autonomous, more agent-like, which is what we're witnessing now. We are in that wave of the Agent Era.
However, there are further questions: Does it have agency? Is it conscious? These are questions we will have to face. But I suggest we take that as step two, perhaps using the tool built in step one to help us explore these deep questions.
Ideally, through this process, we'll also better understand our own brains and minds, and be able to define concepts like "consciousness" more precisely than today.
Host: Do you have any rough predictions about the future definition of consciousness?
Demis Hassabis: No, beyond what philosophy has discussed for millennia, I don't have much to add. But it's clear to me that certain components are obviously required. They might be necessary but not sufficient. Things like self-awareness, the concept of self and other, and some kind of temporal continuity seem clearly necessary for any entity that appears conscious.
However, what the full definition actually is remains an open question. I've discussed this with many great philosophers. A few years ago, I had an in-depth conversation with Daniel Dennett, who sadly passed away recently. One core issue is the system's behavior: does it behave like a conscious system? You could argue that as some AI systems get closer to AGI, they might eventually do that.
But then the question arises: why do we think each other is conscious? Partly because of how we behave; we behave as conscious beings. But another factor is that we are both running on the same underlying substrate.
So I think if both hold, then assuming you and I have similar experiences is logically most parsimonious, which is why we don't usually argue about each other's consciousness. But obviously, we can never achieve the same substrate equivalence with an artificial system. So I think bridging that gap completely is very difficult. You can look at it behaviorally, but experientially? Perhaps there will be ways to address that after achieving AGI, but that might go beyond today's discussion, even in an "AI and Science" conversation.
Host: Excellent. We'll open to audience questions shortly, please prepare your questions. You mentioned philosophers earlier, particularly Kant and Spinoza, as two of your favorites. Kant is a classic deontological philosopher, extremely focused on duty; Spinoza had an almost deterministic view of the universe. How do you reconcile these two very different ideas? What is your fundamental understanding of how the world operates?
Demis Hassabis: The reason I like these two philosophers and am impressed by them is that Kant proposed an idea—something I deeply felt during my neuroscience PhD—that "the mind creates reality," which I think is largely correct. This gives another great reason to study how the mind and brain work. Since I'm ultimately exploring the nature of reality, we must first understand how the mind interprets reality. That's the insight I get from Kant.
As for Spinoza, it's more about the spiritual dimension. If you try to use science as a tool to understand the universe, you start touching upon the deep mysteries behind how the universe operates.
That's what I feel about our current endeavor. When I engage in scientific research, delve into AI, and build these tools, I feel we are, in a way, reading the language of the universe.
Host: Beautiful. That's the most beautiful description of your daily work: Demis, you are a scientist, a speaker, a philosopher. Before we finish, let's do a few rapid-fire questions. He hasn't seen these beforehand. Predict the year for achieving AGI—sooner or later than expected? Or you can decline.
Demis Hassabis: I'll go with 2030. I've been consistent on that prediction.
Host: Okay, 2030. When we achieve AGI, what book, poem, or paper do you recommend as a must-read?
Demis Hassabis: My favorite book for the post-AGI world is David Deutsch's "The Fabric of Reality." I think the ideas there still apply. I'd hope to use AGI to answer the deep questions posed in that book, and that would be my focus of work post-AGI.
Host: Great. Your proudest moment at DeepMind so far?
Demis Hassabis: We've been fortunate to have many high points. I think the proudest is probably AlphaFold.
Host: Okay, final game-related questions. If you were playing a high-stakes turn-based strategy game like Civilization, Polytopia, those hardcore games, and could pick a scientist from history as a teammate, like Einstein, Turing, or Newton, who would you choose for your squad?
Demis Hassabis: I think I'd choose von Neumann. You need a game theory expert in that situation, and I think he's the best.
Host: That would be a god-tier teammate. Demis, you're such a renaissance person. Thank you so much for being here today. Please join me in thanking Demis. Thank you very much.
