© The Financial Times Ltd 2014 FT and 'Financial Times' are trademarks of The Financial Times Ltd.
Last updated: January 27, 2014 4:35 pm
DeepMind, which counted Peter Thiel’s Founders Fund and Li Ka-shing’s Horizons Ventures among its backers, was operating in “stealth mode” and had not launched any products when Google swooped to snap up its team and “machine learning” technology, through which computers teach themselves without human input.
Two people familiar with the situation confirmed the deal. One said Google had moved “extremely quickly” to beat Facebook to the acquisition. A Google insider said the company was paying £400m for DeepMind. Google declined to comment on its purchase.
The FT’s Robert Cookson explores the origin and development of artificial intelligence
“This is my attempt to change the world for the better,” said Mustafa Suleyman, co-founder of the company along with neuroscientist Demis Hassabis and researcher Shane Legg.
Late last year Facebook created an artificial intelligence team to understand the emotions or behaviour implied in posts to its site. IBM is investing $1bn to build a business unit around Watson, its “cognitive computing” project.
DeepMind, which caused a buzz last year when it was presented to tech entrepreneurs and investors at the Founders Forum event in New York, says on its website it aims to “combine the best techniques from machine learning and systems neuroscience to build powerful general-purpose learning algorithms”.
It had said the first commercial products based on its technology would be in games, ecommerce and simulations, but those products are not likely to come to market in their current form after the Google deal.
Little is known about the group, which despite its prominent investors and aggressive hiring spree had kept a low profile in London’s sociable start-up scene. Job advertisements posted online in the past year described it as building “general-purpose learning algorithms”.
Most so-called artificial intelligence algorithms relied on the use of pre-coded “labels” attached to bundles of information, Mr Suleyman said, which required human programmers to be effective.
What do you get if you combine the insights of a neuroscientist, an artificial intelligence programmer, a computer games designer and child chess prodigy?
One answer is a company that, without having released a single product, has become Google’s biggest ever European acquisition . . .
What was distinctive about DeepMind’s technology is that you could set it loose on masses of unstructured data – ranging from web pages to sound and video files – and it would learn about their identity and the connections between them to generate “deep insights”, said Mr Suleyman
Potential applications include a book recommendation engine that draws on a user’s browsing history, and a way of analysing weather data from the internet to produce predictions about climate change, according to Mr Suleyman.
Larry Wasserman, a professor in the machine learning department at Carnegie Mellon University, wrote in a blogpost in 2012 after meeting the company that it was “trying to build a system that thinks”, by combining machine learning and neuroscience.
“I thought it sounded crazy until [co-founder Mr Legg] told me the list of famous billionaires who have invested in the company,” Prof Wasserman wrote.
Given enough computing horsepower, the technology could create a “world of super-intelligent computers surpassing humans”, he speculated.
All the latest news on deals and dealmaking from around the world
One ad, for an intern to work on a “big data” project, began its pitch with: “Are you worried that the only place for smart people with a passion for software in London is in soul-crushing finance? Are you looking to work in a company that invests in hard research to create cutting edge new machine learning algorithms?”
According to Companies House records, DeepMind counts among its directors Bart Swanson, formerly of dating app Badoo and an investor at Horizons Ventures, and Luke Nosek, the PayPal co-founder who formed Founder’s Fund with Mr Thiel.
The company was founded by 37-year-old Demis Hassabis, a London-born teenage chess prodigy who went on to program popular simulation games such as Theme Park. He left gaming to study for a PhD in neuroscience at University College London and published research on amnesia before founding DeepMind in 2012.
The deal is the latest in a growing number of Google acquisitions in futuristic technologies, including robotics pioneer Boston Dynamics, maker of realistic animal and humanoid machines.
This month Google paid $3.2bn for Nest Labs, which makes “smart home” devices, its largest purchase of a private company.
Other Silicon Valley groups are pursuing artificial intelligence technology.
Tech news sites Re/code and The Information earlier reported details of the deal.
Vividly imagined in the science fiction film The Terminator, artificial intelligence is the intelligence exhibited by machines or software.
Ever since the birth of modern computing in the 1950s, futurists have predicted the rise of machines that can think in the same way as human beings, writes Robert Cookson, the FT’s digital media correspondent.
How is it used currently?
So far AI has been limited to solving narrow problems – for example guiding long-range missiles or producing Google’s search results. While these programmes are powerful, they are unable to recognise or respond to new environments the way that people can.
What’s different about DeepMind’s technology?
The founders of DeepMind have backgrounds as neuroscientists. They want to use the principles of how the brain functions to develop sophisticated machine intelligence.
DeepMind has already developed a “neural network” that learned to play seven different Atari 2600 computer games with more skill than a human, using only raw pixels as input and no game-specific information.
The idea is to create software that can learn and respond to messy, unstructured data as found in the real world, rather than applying a rigid set of instructions to pre-existing categories.
What are the potential applications?
In theory AI can be applied to anything. DeepMind has looked at applications including image recognition and book recommendations. But it has much bigger ambitions in robotics.
Shane Legg, one of the founders of DeepMind, has predicted that “human level” artificial general intelligence (AGI) will arrive by 2030. By the 2020, he forecasts the creation of “a system with basic vision, basic sound processing, basic movement control, and basic language abilities, with all of these things being essentially learnt rather than preprogrammed”.
Are we getting a bit ahead of ourselves?
Yes. The brain is a network of 100bn neurons and 1,000 times as many synapses, all working in parallel. Most neuroscientists say a system of this complexity cannot possibly be rivalled by software any time soon.
“If you want to develop systems that have the behavioural flexibility of a human being, there is no option other than to try to recreate the brain’s dynamics inside a computer,” says Simon Stringer, director of the Oxford Centre for Theoretical Neuroscience and Artificial Intelligence, one of the leading laboratories in the field.
Within the next decade, he said, “I would be very pleased if we could develop a system that has the intelligence of a rat.”
Copyright The Financial Times Limited 2014. You may share using our article tools.
Please don't cut articles from FT.com and redistribute by email or post to the web.