My last post was about one of the key figures in Chapter 5, Stephen Hawking, and his terrifying vision of AI making humanity obsolete. The other main character is Demis Hassabis, the founder of DeepMind, and one of the most brilliant British scientists of his generation.
He is pictured above at the tech networking event Founders’ Forum in 2014, just after DeepMind had been sold to Google for something like £400m. That deal had shocked many in the UK’s AI research community who confessed they had never heard of DeepMind and has since provoked plenty of soul searching - from Dominic Cummings amongst others - about why such a precious asset ended up in the hands of a Silicon Valley giant.
This chapter goes into Demis Hassabis’s back story, and explains how an obsession with games - playing them, developing them, even starting his own games company - was central to his thinking about AI. One of his oldest friends, pictured with him above, is Ian Livingstone, the games industry veteran often described as the father of Lara Croft.
He first met Hassabis at a board games evening he hosted, and describes how charismatic he found him, “not only bright, but charming”, words you are unlikely to hear about many figures in the tech world.
But even to get DeepMind off the ground, he had to turn to Silicon Valley, as this extract from the book describes:
So by 2010, he had developed some understanding of how the brain worked, his work in games development had helped him think about how to simulate human interactions, and he had suffered his first failure as a businessman. Just about the perfect grounding then to start his own AI business.
All he and his friends Shane Legg and Mustafa Suleyman needed was a serious chunk of money from investors willing to fund blue-sky research that might not produce a payback for a decade. That was going to be impossible to find from conventional venture capitalists, with AI still provoking, in Demis’s words, ‘eye-rolling’ even in academia. What was needed, he decided, was a billionaire with a passion for this area of science. Where might they be found? Obviously in Silicon Valley, but the three of them had no contacts out there. Then they found out about an event held in San Francisco by the Singularity University, an institution dedicated to the idea that the moment when artificial general intelligence arrives and humans are overtaken by machines is within reach. One of their target billionaires was going to be at the event.
Hassabis managed to get an invitation to speak, and his presentation entitled ‘A systems neuroscience approach to building AGI’, laying out the big advances in understanding the brain and what they implied for artificial intelligence research, was well received. But that was not the point – he still needed to meet his billionaire. A cocktail reception at the conference would provide an opportunity, but he knew that the man in question would be besieged by supplicants just like him, and getting his attention would involve something special.
On an in-house podcast produced by DeepMind Hassabis tells the story of how, once again, a game came to his rescue. He had researched his target and found that they had a common passion for chess.
So I used my number-one fact on chess, which even surprises grandmasters, that, thinking about it from a games designer’s point of view – why is chess such a great game? how did it evolve into such a great game? – my belief is that it is actually because of the creative tension between the bishop and the knight.
In his minute with the billionaire he explained that, while the bishop and knight were worth the same, their different powers produced a creative asymmetry that made chess a fascinating game. ‘I don’t know how I managed to crowbar that into a drinks party, but it made him stop and think, which is exactly what I was hoping.’
The next day Hassabis and his team were invited back to pitch their business idea, and this investor and a couple of others agreed to supply the few million dollars needed to get DeepMind started.
So who was the mystery billionaire? Demis Hassabis has never confirmed the name, but it is widely assumed to be Peter Thiel, the PayPal co-founder and Facebook investor who later became just about the only Silicon Valley figure close to Donald Trump. Even in 2020, when I said, ‘Come on – it was Peter Thiel!’ Hassabis laughed and said if I had got there by a process of triangulation, so be it, but he could not confirm it.
The funding meant they could start work, recruiting some brilliant people who back then, before the AI boom, were still relatively cheap to hire. They were confident that they had the right approach for what was a ridiculously ambitious goal. Their mission statement talked of ‘solving intelligence’, which sounds as though they thought they were charting a clear path to the artificial general intelligence which so frightened Stephen Hawking. But they still had to prove it. ‘We were definitely at the front of the pack, but we were not sure it was going to work.’
To find out how DeepMind became a global leader in AI and what happened when Demis Hassabis met Stephen Hawking to try to set his mind at rest about the threat to humanity you will have to buy the book.
Always On is available as a hardback, ebook or audiobook here.
And if you want to support your local independent book shop you can order it at Hive.