To meet Sir Demis Hassabis is to encounter someone both humble and humbling.
Why humbling? Because Hassabis, 48, is a tech bro, self-made, super-rich, a veritable god in software coding and AI. Humble, because he used his ability to give the world the solution to one of the toughest medical challenges, if not the toughest: to predict the folding of protein into the different shapes that herald all manner of conditions, from Alzheimer’s to cancer.
As a result of his application of AI and his generosity, medical research teams across the planet are accelerating their efforts in leaps that were never thought possible. Breakthroughs are coming and they will be down to Hassabis and his colleague John Jumper, an achievement for which they were jointly awarded last year's Nobel Prize for Chemistry.
This week sees the cinema release of the new documentary film about his life, The Thinking Game. Watching the premiere in London, there are at least three scenes that blow the mind. One is when, early in his AI career, Hassabis programmes a computer to play Pong, the simplest and first video game, the one in which two players hit a ball across the screen until one misses and the other triumphs. Hassabis issues a basic instruction, "work out how to win". For the first six games, the machine loses. On the seventh it wins and after that, never loses. It’s worked out how to win.
Later, he does the same with Go, writing the algorithm that allows the computer to defeat the former world champion in the tactical board game that has far more permutations than chess. To the consternation of South Korea, where Go is hugely popular, the home-grown ex-champ and superstar, Lee Sodol, is beaten.
Hassabis could easily have stopped there, his reputation and fortune – courtesy of DeepMind, his business, being acquired by Google – well and truly made. He chose not to. With Jumper, he went on to develop an AI tool, AlphaFold, that can predict the 3D structure of a protein from its 1D amino acid make-up. Meanwhile, Hassabis maintains his interest in gaming – he was a chess prodigy, the second-best player in the world for his age at 13 – so last year, for good measure, DeepMind’s robotics engineers came up with a system that would win, not at Pong but at actual table tennis.
It is Hassabis’s demeanour and approach that sets him apart. He prefers to focus on AGI, artificial general intelligence, rather than AI per se. His objective is developing AI such that it can overcome not one specific problem but any given task. He could have programmed the computer to win at Pong, it was relatively straightforward, even back then. Instead, he told it to learn for itself how to win.
Similarly, he witnessed Deep Blue defeat the world chess champion, Garry Kasparov, in 1997. While that captivated him, it was not enough – Deep Blue was a chess computer, he wanted something that could not only win at chess but could do much, much more to aid and advance humanity.
He’s feted by world leaders, governments, multinationals, tech billionaires and Silicon Valley. But he eschews the Valley, preferring to stay near where he grew up in North London. His reasoning is that other places are perfectly capable of producing talent and possessing creativity – tech does not have to be concentrated in one area, it’s a global industry. "AI is going to affect the whole world; it’s important there are other voices and cultures involved, not just 100 square miles of California," he has said.
Listening to him in the film, at the Q&A, and meeting him afterwards, it’s impossible not to be inspired. AGI is advancing all the time but is not there yet, despite the best efforts of Hassabis and others. He’s confident, though, that AI will not only show what the proteins causing disease are like but it will soon be able to cure those conditions. To see the myriad molecular structures forming on the screen, thanks to AI, is incredible. Indeed, he has set up a company, Isomorphic, to reach that goal. "I know that sounds crazy today," he has said, "but 10 years ago it would have been crazy for me to tell you we could fold all 200 million protein structures."
This, though, is where he injects a note of caution, if not alarm. While you are left fully appreciating the genuine societal benefits AI will bring, in stark contrast to the constant negativity about threats to jobs and making work tasks quicker, he also wants people to understand where it could easily go so terribly wrong.
He compares where we are now with how, when he was setting out in computing, we were using giant mainframes and the whole business was costly and slow, and only available to the privileged few. In no time at all, computers where on desks, then they were laptops and hand-held. Programming went from being exclusive and rarefied to being practised by people in their bedrooms.
That leap will occur again with AI – it’s already happening – and it worries him. As he puts it, we should be concerned about "bad actors", be they state-sponsored or people acting alone, writing algorithms to cause havoc and untold harm.
We’re receiving regular wake-up calls, yet nothing is being done. One was DeepSeek. The Chinese AI company’s recent unveiling of its own large language model provoked widespread shock. The real lesson, he says, was not so much that China was catching up fast with the US, but that DeepSeek was using much less power at much lower cost. It was a step-change of the sort that suggests AI will soon be modelled in those same bedrooms. What then? Something of colossal capability will be readily accessible by anyone who puts their mind to it.
Governments must act and fast. By all means, hold AI summits devoted to seeking investment, but more urgently we need government-to-government conferences discussing regulation and enforcement. The genie will be out of the bottle and it cannot be put back.
It’s not cheerful, although Hassabis does remain an optimist, with caveats attached. He should be heard. His track record speaks for itself; we must listen.
