Whenever an American VC investor tells you that European founders lack ambition, remind them a London-based startup might yet become the most consequential company in human history.
That company is DeepMind, which has as good a shot as any AI research lab of being the first to develop artificial general intelligence (yes, I know there’s a big dispute about that term but let’s accept it means generalised human-level intelligence for now).
Founded in 2010 by Sir Demis Hassabis, Shane Legg and Mustafa Suleyman, DeepMind has the explicit corporate goal of attaining AGI. Its early mission statement to “solve intelligence” and then “use it to solve everything else” must surely rank as the most crazily ambitious slide ever to appear in a pitch deck.
A great new film, The Thinking Game, tells the compelling tale of what happened over the following 14 years. It colourfully recounts the company’s early hit-and-miss attempts to master the basic computer game of Pong and how it used that experience to tackle the increasingly complex challenges of chess, Go and Starcraft.
The team’s growing expertise in reinforcement learning techniques enabled DeepMind to develop its AlphaFold model to predict 200m protein structures. Historically, it had taken a researcher about five years to model just one protein structure, meaning AlphaFold saved about “1bn hours of PhD time,” according to Hassabis.
The development of AlphaFold, now used by 2.5m biologists around the world, helped Hassabis and fellow DeepMind director John Jumper win a Nobel prize last year. AlphaFold’s success has also boosted DeepMind’s confidence that it is on the right track to develop AGI.
DeepMind is unique in many respects, not least in the exceptional talent of Hassabis himself. But there are perhaps three broader lessons that can be drawn from its experience that are relevant to all European founders and investors.
First, DeepMind combines crystal-clear strategic clarity with never-ending tactical flexibility. What comes across in the film is the company’s extraordinary willingness to experiment wildly and fail persistently. The movie audience burst into laughter when Sir Paul Nurse, the Nobel prize winning biologist, appeared on screen to say that much of his time as a scientist was spent consoling colleagues whose experiments had failed. But learning from failure is at the core of all scientific — and very often corporate — progress.
Second, DeepMind’s mission has helped it recruit some remarkable scientific talent, critical to its success. In a discussion after the movie, Hassabis explained that he had always resisted investor pressure to move to Silicon Valley and had been determined to remain in London. “The UK has always been very strong in science and innovation and has a rich history in computing,” he said. “We are trying to carry on in that tradition.”
Hassabis reckoned that there was a lot of under-utilised academic talent in Europe, and elsewhere, that could be attracted to London. So it has proved.
Third, what was essential for DeepMind’s success was its ability to scale rapidly. Back in 2010, few VCs were prepared to go anywhere near a startup with such extravagant ambitions and no business plan. Much of its initial capital came from US investors, including Peter Thiel and Elon Musk. The company also felt compelled to sell out to Google in 2014 to give it the capital, data and computing firepower necessary to stay at the leading-edge of AI. (The extra resources were essential for recruiting and retaining top talent, too).
The European startup ecosystem has evolved a lot in the past 15 years and there are now far more founders-turned-funders — including Daniel Ek, Xavier Niel, Taavet Hinrikus, Niklas Zennstrom and Jaan Tallinn — who are prepared to back madly bold founders. But we now need the big European institutional investors to weigh in if we want more local startups to emulate DeepMind and shoot for the moon.
Read the orginal article: https://sifted.eu/articles/deepmind-europe-startups-learn-john-thornhill/
