Demis Hassabis

In New Delhi this February, on stage at the India AI Impact Summit, Demis Hassabis gave away the whole game in one sentence. DeepMind's founding mission, he said, was "solve intelligence step one and then step two, use it to solve everything else" (ANI, Feb 2026). At the time it "sounded like science fiction." Now he thinks it's roughly on schedule. That is the load-bearing belief: not that AI is useful, but that intelligence itself is a solvable problem, and that solving it dissolves every other problem downstream. Everyone else in this field sells products. Hassabis is selling a theory of everything.

What makes him singular is that he can almost point to the proof. AlphaFold predicted the structures of 200 million proteins, is used by over 2 million researchers, and has been cited more than 30,000 times (Mad SciHub, Feb 2026). He doesn't argue AI will be good for science. He shows you the receipt and dares you to find another lab leader holding one.

The clock he set himself

Hassabis is precise about timelines in a way that should make you nervous. He told the Delhi summit that DeepMind, founded in 2010, conceived of a "20-year mission" and is "basically on track for that, you know, around 2030" (ANI, Feb 2026). In his keynote he tightened it: AGI is "on the horizon maybe in the next five, 5 to 8 years."

But press him on whether we're "there yet" and the optimist becomes a tough grader. "No. I don't think we're there yet," he said, citing a high bar: a system showing "all the cognitive capabilities humans can, including creativity, long-term planning." Today's models, he argues, are "jagged intelligences" ... they can win gold medals in the International Math Olympiad yet "make mistakes on elementary maths if you pose the question in a certain way" (ANI, Feb 2026). What's missing: continual learning, long-horizon planning, consistency, and the thing he keeps circling back to, taste.

His test for real intelligence is gorgeous and brutal. Train a model with a knowledge cutoff of 1911, he proposed, and see if it produces general relativity the way Einstein did in 1915. "Today's systems clearly would not be capable of doing that." So here's the tension he carries: AGI is five to eight years out, and also the systems can't reliably do arithmetic. Both are true. He believes the gap closes anyway.

The two worries he won't soften

For a man whose default register is sunny ("the next 10 years it'll be almost like having superpowers"), Hassabis is unusually blunt about danger. He names two risks, repeatedly, almost as a liturgy: bad actors repurposing dual-use systems ("individuals, but could be also nation states") and the technical problem of control as models go agentic (BBC, Feb 2026).

On the first he gets specific in a way most CEOs avoid: "we need to worry about things like bio and cyber risk very soon... you need to make sure the defenses are stronger than the offenses" (ANI, Feb 2026). On the second, the honesty is striking from someone whose product roadmap depends on the optimistic answer. Asked how to build guard rails for autonomous systems, he said flatly: "I think there's still an open research question and I think more research needs to be done urgently" (BBC, Feb 2026). He told the BBC the industry "wanted smart regulation."

Then comes the hedge that holds his whole worldview together: "I'm a big believer in human ingenuity. I think there are many ways it could work but we need to make sure it does work." Read that twice. The man building the autonomous systems is telling you the safety problem is unsolved, the agentic era arrives "this year and next year," and his reassurance is essentially a vote of confidence in our species. It's faith, dressed as a research program.

Curing disease at "digital speed"

Where Hassabis stops hedging and starts evangelising is medicine. AlphaFold, he insists, "was just the beginning" ... a 50-year grand challenge that turned out to be one small (if important) piece of drug discovery (IISc, Feb 2026). Through the spinout Isomorphic Labs, the goal is to compress drug development from "an average of 10 years" to "a matter of months, maybe even possibly weeks" (Mad SciHub, Feb 2026). His framing: "doing science at digital speed." His dream: a virtual cell you can run experiments on in silico.

He frames this almost theologically. "I think AI is potentially the perfect description language for biology," he said, the way maths was for physics (Mad SciHub, Feb 2026). Biology, to him, is "an information processing system that's trying to resist entropy" ... which is either the most profound or most reductive thing anyone has said about life this year, depending on your seat.

And here, notice, the safety caution doesn't vanish, it gets operationalised: DeepMind "consulted with over 30 biosecurity and bioethics experts" before releasing AlphaFold (Mad SciHub, Feb 2026). The worry and the shipping coexist. That's the Hassabis method.

The polymath alibi

To understand why one person runs a frontier lab, a drug-discovery company, and a games world-model project, you have to take the polymathy seriously, because he does. He decided "around about 16 years old, that AI was going to be my career" while working at the games studio Bullfrog (IISc, Feb 2026). His heroes are "da Vinci or Aristotle, who didn't really see the boundaries" between art, science and philosophy.

The tell is in how he organises companies: DeepMind as "a combination of neuroscience, engineering and machine learning," Isomorphic as "an intersection of machine learning, chemistry and biology." And the confession underneath it: building AI as the "ultimate tool for science" has "given me the excuse to learn about a lot of other subject areas" (IISc, Feb 2026). The whole enterprise is, partly, one extremely curious man's permission slip to be curious about everything at once. He has, plausibly, built the most powerful research instrument in history because he could not pick a single subject.

So watch what he actually said, not the inspirational version. The race is on, the timeline is set, the control problem is unsolved, and the man steering has wagered the species on human ingenuity arriving before AGI does. He's not wrong that he's been right before. He's just never been right about something this large, on a clock he started himself.