Solve intelligence, then everything else: Demis Hassabis

The pitch sounded, by any reasonable measure, like hubris. In 2010, three founders started a small London research lab and told prospective backers what they intended to do with it: first they would solve intelligence itself, and then they would use that intelligence to solve everything else. Not a product. Not a market. Intelligence, the general-purpose capacity to learn, treated as a single engineering problem, and then pointed at every other problem in turn.

One of those three founders was Demis Hassabis, and he had a name for the ambition. DeepMind, he said, was conceived as "the Apollo programme for AI", a focused, mission-driven push at one grand challenge, the way NASA had pointed itself at the Moon. It is the kind of line that ages badly if the work never arrives. Fourteen years later, the work arrived in Stockholm: in 2024 Hassabis shared the Nobel Prize in Chemistry for an AI system that had cracked a fifty-year-old problem in biology. The pitch, it turned out, had been a forecast.

The chessboard, the arcade, the brain

To understand the bet, it helps to see how unusually Hassabis arrived at it. Born in London in 1976, he learned chess at four by watching his father play, and was a child prodigy at the board, reaching master standard at thirteen, captaining England junior teams, and at one point ranked among the strongest under-14 players in the world. Chess is where he first met the idea that runs through everything since: that intelligence is something you can study, decompose and improve.

He did not stay at the board. As a teenager he moved into games, and at seventeen he co-designed and was lead programmer on Theme Park (1994), a business-simulation game made under designer Peter Molyneux at Bullfrog Productions. It was a genuine hit, it sold into the millions and helped seed an entire genre of "sim" games, and it taught him a second lesson the lab would later run on: that you can build a system that learns from and adapts to the person playing it.

Then came the most deliberate move of all. Rather than ride the games success, Hassabis went back to first principles, taking a PhD in cognitive neuroscience at University College London (completed 2009), studying memory and imagination in the human brain. The logic was almost startlingly literal: if you want to build intelligence, first go and study the one working example nature has already produced. Chess, games and neuroscience were not three careers. They were one continuous question, what is intelligence, and could you rebuild it, approached from three directions.

An Apollo programme, and a Go board

DeepMind, founded in 2010 with Shane Legg and Mustafa Suleyman, was the synthesis. Google acquired it in 2014, reportedly for around £400 million, and, crucially, let it keep operating as a research mission rather than a product unit. That distinction matters to how Hassabis leads: he is the scientist-CEO who still frames the organisation as a lab with a thesis, not a business with a roadmap.

"DeepMind is the Apollo programme for AI."

The thesis announced itself to the world in March 2016, when DeepMind's AlphaGo defeated Lee Sedol, one of the greatest Go players alive, 4–1 in Seoul. Go had long been considered beyond the reach of machines, too vast and too intuitive for brute-force search, and the win was widely read as a milestone moment for AI. But for Hassabis a game-playing system was never the destination. It was a proving ground. Master a closed world like Go, and you have built learning machinery you can then aim at the open, messier worlds that actually matter.

The thesis, made real

The world that mattered most turned out to be biology. Proteins fold into elaborate three-dimensional shapes that determine what they do, and predicting that shape from a sequence of amino acids had defied scientists for roughly half a century. In late 2020 DeepMind's AlphaFold was judged to have all but solved it; the team then released predicted structures for nearly every protein known to science, and the database has since been drawn on by millions of researchers in fields from drug discovery to neglected diseases.

In October 2024 the work was recognised at the highest level in science: Hassabis and his DeepMind colleague John Jumper were awarded the Nobel Prize in Chemistry, shared with the biochemist David Baker, for protein-structure prediction. Think about how odd that is. A computer scientist who had spent his career insisting that AI was a route to scientific discovery was handed a Nobel in chemistry for doing exactly that. The same year he was knighted for services to artificial intelligence, having earlier been made a CBE in the 2018 New Year Honours. In his Nobel lecture that December he gave the through-line a plain title: Accelerating scientific discovery with AI.

Intelligence as a meta-tool

What separates Hassabis from most of the people now building large AI systems is the order of his reasoning. The fashionable framing treats AI as a product to be shipped. His framing treats intelligence as a meta-tool, the one capability that, once built, unlocks every other problem downstream. AlphaFold is the clean demonstration: you do not solve protein folding by being a better biologist; you solve it by building a better general method and then aiming it at biology. The same logic now points at drug discovery through Isomorphic Labs, the DeepMind spin-out he also leads, where the wager is that the structure-prediction breakthrough can be turned into actual medicines.

That order of reasoning is also a caution against reading the Nobel as the finish line. By his own framing the protein work is a proof of concept for a much larger claim, that AI can compress the timescale of scientific discovery itself across disciplines. It is a thesis with obvious risk attached, and Hassabis has been notably willing to talk about the dangers of the technology as well as its promise, sitting on government advisory roles and arguing for caution even as he accelerates the work.

There is a tidy irony in where he has landed. The four-year-old at the chessboard, the seventeen-year-old programming a theme park, the doctoral student mapping human memory, each was, in retrospect, studying the same thing. The Apollo line could have been the epitaph of an overreaching lab. Instead it became a description of what the lab did. The pitch was always two steps: solve intelligence, then use it to solve everything else. The Nobel was the world conceding that the first step is no longer fantasy, which makes the second step, the audacious one, the part still being written.