Demis Hassabis, the Nobel winner who stepped back from Google’s AI race to call for a referee

Demis Hassabis

Photo: John Sears / CC BY-SA 4.0, via Wikimedia Commons

Born	July 27, 1976 London
Occupation	Engineer
Awards	Mullard Award · Commander of the Order of the British Empire · Princess of Asturias Award for Technical and Scientific Research

For most of his adult life Demis Hassabis argued that the safest way to build a machine smarter than people was to build it himself, inside a lab that answered to science before it answered to a product calendar. Now he has handed the daily running of that lab to someone else. The man who turned Google DeepMind into the engine behind Gemini is its chair, Alphabet’s chief scientist and the head of a drug-discovery company that has yet to put a medicine into a patient. His note to staff was blunt: artificial general intelligence, he wrote, is close at hand, and getting the next steps right is critical for humanity. The builder has concluded that the most useful thing he can do is help write the rules.

Hassabis has treated intelligence as a game to be solved for as long as anyone has known him. He was born in London to a Greek Cypriot father and a Chinese Singaporean mother, and was playing chess at four. By thirteen he had reached master standard and was captaining England’s junior teams. Prize money from the board paid for a ZX Spectrum, and the computer became his second obsession.

At 17, having finished his A-levels early, he co-designed and lead-programmed Theme Park with Peter Molyneux at Bullfrog Productions, a simulation in which players build and run an amusement park and which sold millions of copies. He then read computer science at Queens’ College, Cambridge, graduating with a double first in 1997, worked as lead AI programmer on Molyneux’s Black & White at Lionhead, and in 1998 founded his own London studio, Elixir. Its two releases, the political simulation Republic: The Revolution and the villain-lair comedy Evil Genius, drew respectable rather than rapturous reviews, and Elixir closed in 2005.

Instead of starting another company, he went back to university. At University College London, under the neuroscientist Eleanor Maguire, he studied patients with damage to the hippocampus and found they struggled to imagine new experiences, evidence that memory and imagination share machinery. Science listed the work among the ten breakthroughs of 2007, and he completed his PhD in cognitive neuroscience in 2009 before postdoctoral spells at MIT and Harvard and a fellowship at UCL’s Gatsby unit. The detour was the plan: study the only general intelligence available, then try to build one.

DeepMind, founded in London in 2010 with Shane Legg and Mustafa Suleyman, set out to solve intelligence and then use it on everything else. Google bought it in 2014 for around £400 million. In March 2016 its AlphaGo program beat Lee Sedol 4-1 in Seoul. The bigger bet came next. In November 2020 AlphaFold predicted protein structures with near-atomic accuracy in the CASP14 assessment, and DeepMind later released, with EMBL’s European Bioinformatics Institute, predicted structures for some 200 million proteins, free for any lab to use. That work earned Hassabis and his colleague John Jumper a share of the 2024 Nobel Prize in Chemistry, alongside David Baker. He was knighted the same year. In 2023 Google merged DeepMind with its Brain team and put Hassabis in charge of the combined Google DeepMind, which builds the Gemini models.

The scientist’s halo has never covered everything. In 2017 the UK Information Commissioner ruled that the Royal Free NHS trust had failed to comply with data protection law when it handed records of about 1.6 million patients to DeepMind to test Streams, an alert app for acute kidney injury. No fine followed, but the case showed that a lab that sold itself on scientific seriousness could chase data like any other technology company. Isomorphic Labs, the Alphabet spin-off he founded in 2021 to turn AlphaFold into medicines, had been expected to reach clinical trials by the end of 2025; at Davos in January he moved that to the end of 2026 and told Bloomberg he had misspoken, meaning pre-clinical work. A $2.1 billion funding round led by Thrive Capital followed in May, yet by late September no patient had been reported dosed with an Isomorphic compound. And the August reshuffle landed while Fortune was reporting that Gemini’s next flagship was running months late; Alphabet shares fell about 5% that day. The Nobel has not settled whether he is a scientist who happens to run a business or an executive carrying a scientist’s reputation.

His recent moves read as an answer of sorts. In July he published a proposal for a Frontier AI Standards Body modelled on FINRA, the American financial industry’s self-regulator: funded by the AI companies, overseen by government, and testing the most capable models for cyber, biological and deception risks up to 30 days before release. In August Koray Kavukcuoglu, until then DeepMind’s chief technology officer, took over day-to-day leadership as a senior vice-president reporting to Sundar Pichai, while Jeff Dean left Google after 27 years. When Dario Amodei, Anthropic’s chief executive, published his essay “We Must Pace the Frontier” on 12 September, Hassabis replied within hours that it pointed towards the right path forward. On 30 September DeepMind published SynthID Bio in Nature, a method for invisibly watermarking AI-designed proteins, and opened the code and weights to other researchers; provenance for AI-designed biology, Hassabis said, is becoming urgent.

Outside the lab he keeps the habits of a games player. He won the Pentamind, the multi-game title of the Mind Sports Olympiad, five times, and remains a devoted Liverpool supporter. Hassabis, who turned 50 in July, lives in North London with his family. His career has already been told twice: in the documentary The Thinking Game, which premiered at Tribeca in 2024, and in Sebastian Mallaby’s biography The Infinity Machine, written after years of conversations with him.

The next verdict on him will come from a clinic. Isomorphic’s first AI-designed candidates, aimed at cancer and immune disorders, are due to enter human trials before the end of this year, the deadline he has already moved once. If they start on time, the claim he has made since his PhD, that understanding intelligence is the shortest route to everything else, gets its first test in a human body. If they slip again, the argument he is now making to regulators and rivals about how fast this technology should move will be heard from a man whose own timetable keeps running late.

Tags: business, ceos, business leaders, Demis Hassabis, Google DeepMind, Koray Kavukcuoglu