Demis Hassabis on Why AGI Will Deliver Radical Abundance—And What Could Go Wrong
If you accept that artificial intelligence is the most transformative technological shift humanity will ever encounter, Demis Hassabis’ views deserve your full attention. As the leader of Google’s global AI efforts, he helms one of the best-resourced teams in the world, backed by billions in investment to pull off this unprecedented revolution. He’s also among the most prominent powerful tech leaders racing to build artificial general intelligence (AGI)—the holy grail of AI that would let machines match, and outperform, nearly every cognitive task humans can do.
Unlike nearly every other leader in this cutthroat race, Hassabis boasts both a Nobel Prize and a knighthood earned for his work—all of it rooted in his lifelong love of games. Raised in London, he was a teenage chess prodigy, ranked second globally among players his age at just 13. He was also obsessed with complex computer games, first as a top-tier player, then as a designer and programmer behind beloved classic titles like Theme Park. But his core ambition has always been to build machines as sharp and creative as he is. He left the gaming industry to study the human brain, earning his PhD in cognitive neuroscience in 2009. A year later, he launched his audacious quest: co-founding DeepMind to build AGI from the ground up.
Google acquired DeepMind in 2014, and more recently merged it with Google Brain, Google’s product-focused AI research division, putting Hassabis in charge of the combined group. One of his biggest wins came from applying game-inspired AI methods to the decades-old scientific problem of predicting protein structure from amino acid sequences: AlphaFold, the breakthrough that won him the Nobel Prize in Chemistry last year.
Today, Hassabis is going all in on what may be the biggest game of his career: delivering AGI, amid a ferocious global race with other private companies and Chinese tech firms. On top of that, he serves as CEO of Isomorphic Labs, an Alphabet subsidiary that leverages AlphaFold and other AI advances to accelerate new drug discovery.
When I sat down with Hassabis at Google’s New York City headquarters, his answers came as quickly and smoothly as a top-tier chatbot’s response. He parried every question I posed with good humor and unshakable confidence that he and Google are on the right path. Ask if we need a massive new breakthrough to reach AGI? Yes, but it’s already in development. Ask if more powerful AI brings catastrophic existential risk? Don’t worry—AGI itself will help solve those problems. Ask if it will wipe out most of today’s jobs? Maybe, but there will still be work enough for many. That, remarkably, is his version of optimism. You may not agree with everything Hassabis says, but his ideas and his next moves will shape all our futures. After all, history is written by the winners.
This interview has been edited for length and clarity.
Q: You launched DeepMind with a stated 20-year mission: first crack the problem of intelligence, then use that invention to solve every other major problem facing humanity. You’re 15 years into that timeline—are you still on schedule?
A: We’re almost exactly on track. In the next five to 10 years, I’d say there’s roughly a 50% chance we’ll deliver the version of AGI we’ve been working toward.
Q: What’s your working definition of AGI, and what makes you confident we’re this close to achieving it?
A: There’s a long-running debate over how to define AGI, but we’ve always framed it as a system that can demonstrate every single cognitive capability that humans possess.
Q: Former Google CEO Eric Schmidt has argued that if China develops AGI first, the West is at an unbeatable disadvantage—he argues the first AGI will pull further and further ahead exponentially, building an insurmountable lead. Do you disagree with that take?
A: That outcome is impossible to predict right now. That’s the so-called hard takeoff scenario, where AGI can rapidly iterate on and improve its own code, turning a tiny early lead into a massive gap in just days. My bet is that progress will be far more incremental. It will take more than a decade, likely, for the full impact of advanced digital intelligence to reshape most real-world systems.
Q: Given that hard takeoff is a possible outcome, does Google see being the first to develop AGI as an existential priority?
A: This is an incredibly intense moment for the field, with massive amounts of investment pouring in, huge pressures, and so much research still left to do. Of course we want to unlock all the incredible good that advanced AI can deliver: new cures for disease, breakthrough clean energy sources, transformative advances for all of humanity. But if the first AGI systems are built with misaligned values, or developed without proper safety precautions, the outcome could be devastating.
I see two major risks that keep me up at night. First, bad actors—whether rogue individuals or hostile states—could repurpose AGI for harmful ends. Second, there’s inherent technical risk in the AI itself: as systems get more powerful and more autonomous, can we build safety guardrails that hold up and can’t be bypassed.
Q: Just two years ago, nearly every major AI company including Google was calling for regulation. Today, at least in the U.S., the government seems far more focused on accelerating AI development to outcompete China than putting new rules in place. Do you still support government regulation of AI?
A: Smart, targeted regulation still makes all the sense in the world. It has to be flexible, able to adapt as our understanding of AI research evolves. And most importantly, it has to be global—that’s the biggest challenge we face right now.
Q: If you get to a point where AI progress is moving faster than our ability to make systems safe, would you support a pause in development?
A: Current AI systems don’t pose any existential risk, so this is still a theoretical question. Geopolitical pressures may end up being a trickier challenge than safety itself. But given enough time, care, and rigorous application of the scientific method, I believe we can solve these problems…
Q: If your timeline is right, and we could have AGI in 5 to 10 years, we don’t have a lot of extra time for that care and deliberation.
A: You’re right, we don’t have a lot of time. That’s why we’re already pouring more and more resources into security, from cyber protections to research into AI controllability and mechanistic interpretability—work that helps us understand exactly how these advanced systems work. At the same time, we need broader societal conversations about building the right institutions and governance frameworks. We need to work out how global governance should function, and reach international agreement on at least basic ground rules for how AGI systems are built, deployed, and used.
Q: How extensively do you expect AI to change or eliminate human jobs?
A: Historically, new technologies tend to create new, better jobs that leverage these new tools. Time will tell if this time is different, but over the next few years, AI will act as an incredible productivity supercharger for human workers, almost amplifying our capabilities to superhuman levels.
Q: If AGI can do every cognitive task a human can do, it can do those new jobs too, right?
A: There are plenty of roles that humans will always want to be filled by other humans. AI can support a doctor, or even act as a basic AI care provider, but most people don’t want a robot nurse—human connection and empathy in care are inherently human traits that we will always value.
Q: What does your ideal 20-year future look like, where AGI is fully integrated into everyday life?
A: If everything goes right, we’ll enter an era of radical abundance, a new golden age for humanity. AGI can solve what I call the root-node problems holding us back: curing incurable diseases, extending healthy human lifespans, unlocking affordable clean energy for everyone. If we pull that off, it will be an era of unprecedented human flourishing—one where we can even begin exploring space and working toward interplanetary colonization. I think we’ll start seeing that progress as early as the 2030s.
Q: I’m skeptical. We already have extraordinary abundance in the Western world, but we don’t distribute it fairly. And when it comes to solving big global problems, we don’t lack solutions—we lack the political will to implement them. We already know how to address climate change, for example, we just haven’t done it.
A: I agree with that. As a species and a global society, we’ve never been good at large-scale collaboration. Destroying our natural habitats is happening in large part because it requires short-term sacrifices that people and nations are unwilling to make. But the radical abundance AI delivers will turn every global challenge into a non-zero-sum game—
Q: So you think AGI would actually change how humans behave and cooperate?
A: Yes. Let me give you a simple example. Access to clean water is going to be one of the biggest global crises of this century, and we already have a solution: desalination. It just requires massive amounts of energy. If AI helps us deliver cheap, unlimited, clean fusion energy, desalination becomes accessible to every country, and suddenly the water crisis is solved. It’s no longer a fight over limited resources—a zero-sum game where one side has to lose for the other to win.
Q: If AGI solves these resource problems, do you think it will make humanity less selfish?
A: That’s what I hope. AGI will deliver radical abundance, and then—this is where I think we need great philosophers and social scientists to lead the way—our society as a whole will shift to a non-zero-sum mindset.
Q: Do you believe that leaving this groundbreaking innovation in the hands of for-profit companies is the right approach?
A: So far, capitalism and Western democratic systems have proven to be the most effective drivers of technological progress. Once we reach the post-AGI era of radical abundance, we will need entirely new economic theories, and I’m surprised more economists aren’t already working on that shift.
Q: Whenever I write about AI, I get flooded with messages from people who are deeply angry about it. It reminds me of artisans who lost their livelihoods during the Industrial Revolution—they feel AI is being forced on the public without any input or consent. Have you encountered this anger and pushback?
A: I haven’t encountered much of it personally, but I’ve read and heard plenty about it, and it’s completely understandable. This transformation will be at least as big as the Industrial Revolution, and probably much larger. It’s normal to be scared of that much change.
That said, when I explain to people why I’m building AI—to advance science, medicine, and our understanding of the world—I can point to tangible proof that it’s not just empty talk. AlphaFold is a Nobel Prize-winning breakthrough that’s already accelerating medical research and drug discovery. When people see that, most agree that it would be immoral not to pursue this technology if it’s within our reach. I’d actually be far more worried about our future if we didn’t have a revolution like AI coming to help us tackle our biggest global challenges. AI is a challenge in itself, of course, but if we get it right, it’s the key to solving all our other problems.
Q: You’ve had a lifelong background in gaming—how has that shaped the work you’re doing now in AI?
A: Competing internationally in chess as a kid gave me incredible training for handling pressure, which has been extremely useful in the hyper-competitive AI landscape we work in today.
Q: Games are easier for AI to master because they have clear, fixed rules. We’ve already seen AI pull off shocking, genius-level moves in games—think of Deep Blue’s iconic "Hand of God" move against Kasparov, or AlphaGo’s groundbreaking Move 37 against Lee Sedol. But the real world is far more unstructured and complex. Do you think we’ll see AI make similarly counterintuitive, transformative breakthroughs in real life?
A: That’s exactly the dream.
Q: Do you think AGI can ultimately uncover the fundamental rules that govern reality itself?
A: That’s exactly what I hope AGI will deliver: an entirely new theory of physics. Right now, no AI system can invent a game as deep and elegant as Go. We can use AI to solve math problems, even maybe one of the Millennium Prize problems. But can an AI come up with something as groundbreaking and profound as the Riemann hypothesis? Not yet. That requires true inventive creativity, which current systems don’t have.
Q: It would be extraordinary if AI could crack the fundamental code that underpins the universe.
A: But that’s exactly why I started this work. It’s been my goal since I was a child.
Q: To solve the question of existence itself?
A: Reality—understanding the fundamental nature of reality. It’s right in my Twitter bio: “Trying to understand the fundamental nature of reality.” It’s there for a reason. That’s the deepest question humanity can ask. We still don’t understand the nature of time, or consciousness, or even basic reality itself. I don’t understand why more people don’t focus on these questions—they’re right in front of all of us.
Q: Have you ever taken psychedelics like LSD? That’s how many people say they get a glimpse of the fundamental nature of reality.
A: No, I never have, and I don’t want to. That’s not how I’ve explored those questions. I got that stimulation from gaming and reading constantly as a kid—both science fiction and hard science. I’ve studied too much neuroscience to risk altering my brain, honestly. I’ve spent my life refining how my mind works, and I need it sharp for the work I’m doing.
Q: This is all very profound, but you’re also responsible for leading Google’s cutthroat AI competition right now. Right now it feels like a constant game of leapfrog: every few weeks, either Google or a competitor releases a new model that claims to be the best on some niche benchmark. Is a massive leap coming that will break us out of this incremental cycle?
A: We have the deepest research team in the field, and we’re constantly working on what we internally call the “next transformer” — the next foundational breakthrough that will change everything.
Q: Do you already have promising candidates for a breakthrough as big as transformers, that would deliver another massive jump in AI capability?
A: Yes—we have three or four really promising lines of research that could grow into a leap as big as transformers were.
Q: If you have that breakthrough, how will you avoid repeating the mistake Google made with transformers? Google invented the transformer architecture back in 2017, but failed to capitalize on its advantage—OpenAI ended up building on it first and launching the generative AI boom we’re in now.
A: We’ve definitely learned lessons from that period. Back then, we were too focused on pure research. In hindsight, we should have not only invented the architecture, but moved faster to productionize and scale it. That’s absolutely our plan this time around.
Q: Google is one of many companies working to build AI agents that can complete full tasks for users autonomously. Is the biggest challenge here making sure these agents don’t make costly mistakes when they act on their own?
A: Every leading lab is working on agents because they’re far more useful than current AI tools. Today’s large language models are basically passive question-and-answer systems. You don’t just want your AI to recommend a restaurant—you want it to book the reservation for you, too. But yes, autonomous agents bring new safety challenges, and we’re working extremely hard on building strong guardrails and testing them thoroughly before releasing them to the public.
Q: Will these agents eventually become persistent, always-on companions that handle all sorts of ongoing tasks for you?
A: I call this the universal assistant. Eventually, you’ll have a system that’s so useful you’ll use it every day, all day. It will be a constant companion that knows you, knows your preferences, enriches your life, and makes you far more productive.
Q: Can you explain the new AI Mode Google announced at I/O this year? You added AI Overviews to search that summarize results at the top so users don’t have to click through, and now AI Mode lets users get full answers from a powerful chatbot directly in search. Does this mean Google is moving away from traditional search to a new paradigm where generative AI chat becomes the primary way people access the world’s information? If Gemini can answer all my questions, why would anyone use traditional search anymore?
A: There are clear use cases for both. When you want to get facts quickly, confirm details, and easily check original sources, AI-powered search with AI Overviews works perfectly. If you want to do deeper, more exploratory searching and brainstorming, AI Mode is ideal for that use case.
Q: But we’ve been talking about the future where all interaction with technology is a continuous conversation with an AI assistant. Where does traditional search fit into that future?
A: Steven, let me put it this way: you probably have a human assistant, I’ve had a fantastic human assistant working with me for 10 years. I don’t go to her for every single quick information need I have—I just use search for that, right?
Q: But your human assistant doesn’t have access to all of human knowledge. Gemini aims to have that, so why would anyone need traditional search anymore?
A: All I can say is that right now, and for at least the next two to three years, both modes will be necessary and both will keep growing. We plan to lead in both.
Have thoughts on this interview? Send a letter to the editor to be considered for publication at [email protected].
