Computer that beat Go champion has many applications, says its maker.
A little chalkboard sits on the reception desk of DeepMind's office in London's gentrified King's Cross. On it is scrawled: AlphaGo - 4, Lee Sedol - 1. Nearby in the lobby, a big-screen TV is flashing the words: Welcome back AlphaGo Team!
But that is about as far as one can tell that the London company has just come home triumphant after making history last week by trouncing Go world champion Lee Se Dol with its supercomputer, AlphaGo.
Perhaps the team already knew they were going to win the best-of- five epic showdown between man and machine in Seoul. Engineers had been testing AlphaGo against older versions of itself so they could get an estimate of how strong the machine was - it was very strong.
"Of course we didn't know for sure," said DeepMind co-founder and chief executive Demis Hassabis modestly. "Lee Se Dol is one of the greatest players of all time. He's famous for being very creative in his play, so he might have come up with some strategies that our systems didn't know."
Since the systems learn by playing against themselves, there could potentially be a "blind spot" in their knowledge if no player has ever explored a certain strategy, he said.
AlphaGo stumbled in the fourth round against Mr Lee, who took that game after five hours of play. But the Go grandmaster could not defend the human race against the formidable self-learning machine that swept four of the five games.
Machines have trumped humans before. IBM's Deep Blue took on world chess champion Garry Kasparov nearly 20 years ago and won, through sheer computational power.
But Go is a far more complex and intuitive game than chess, with more possible configurations than there are atoms in the universe - no computer can crunch that.
The souped-up contender this time, AlphaGo, has algorithms modelled after the human brain - it learns on its own, adapts and gets better with practice. The potential applications of such a system to a wide-range of fields are what excite scientists, including Mr Hassabis.
"In the next five years, it would be great to see machine-learning be applied to healthcare in a deep way for medical diagnosis," he said.
Indeed, machines are not something to be feared and will probably never take over the world and wipe out humanity, never mind what the movies tell you.
"There are these science fiction scenarios but they're just science fiction. I don't think we should confuse Hollywood and what's really reality," said Mr Hassabis, smiling.
His vision for artificial intelligence: making it a tool to help experts in anything - from cancer research to climate change - understand their fields better.
Mr Hassabis has often been described as a polymath. He taught himself to write computer programs by the time he was eight; reached chess master level when he was 13; became a successful video game designer by 17; graduated with a double first in computer science from Cambridge University; and then earned a PhD in cognitive neuroscience from University College London.
In 2010, he co-founded DeepMind, a machine-learning start-up that Google paid £400 million (S$770 million) for four years later that now has 250 employees.
The 39-year-old credits Singapore for getting him interested in computers. Born to a Greek Cypriot father and Chinese Singaporean mother - both teachers - in North London, he spent his summers in Singapore up until he was 10.
"Back in the early 80s, there were so many cool gadgets that were coming from Japan that you could buy in Singapore that you couldn't buy in Britain. I remember getting Nintendo Game and Watch's Donkey Kong. That was my favourite," he said.
He hopes to visit Singapore soon - the last time he was there was more than two years ago.
"When I was a kid, I'd always think of Singapore as this magical futuristic world. And I think it's still a bit like that today. I'd like to go back there more often," he said.
What would it take, then, for the city-state to successfully woo a top mind like him?
"You need to create enough of a critical mass of people that you can build a company like DeepMind out of," he said, adding that Britain's collection of top-class universities means talent is being trained here. "You probably got enough of the ingredients. You got money, government will, and very smart people and top universities and it's a highly technological society. You may just need the right entrepreneur or leader to pull it all together."
If there was one big lesson he took away from the Seoul experience, it was realising how fast things can progress if they are organised right. The AlphaGo project started less than two years ago with a core team of three people. More came on board as time went on, but there was never more than a dozen at any time.
"From start to finish in two years and beating the world champion is pretty amazing, I think," he said.
The team is considering releasing AlphaGo in some form. The Go community, for one, would relish that, since players can train against this current world champion.
"But it's a lot of work to package something that's very prototype into something that could be released, so we have to decide the feasibility of that."
This article was first published on March 25, 2016.
Get a copy of The Straits Times or go to straitstimes.com for more stories.