How Google's new AI genius changed gaming

Why Google will be playing mind games with customers - courtesy of Theme Park.

There’s big news out of Silicon Valley this morning, with search giant Google announcing it’s picked up artificial intelligence (AI) start-up DeepMind Technologies for reportedly more than $500m (£302m).

DeepMind Technologies logo
DeepMind© DeepMind Technologies

The London-based start-up has around 75 employees, and will now put its talented hive mind to use in helping Google either fend off or usher in Skynet, depending on how much of a technological optimist you are.

But what you might not have realised is that one of its founders, neuroscientist Demis Hassabis, first cut his teeth in the field of AI working on some of Britain’s best games in the 90s and 00s - and many of those games hinged upon the quality of the AI.

Hassabis is a neuroscientist who has spent years studying how human autobiographical memory works, but he wasn’t always an academic. Hassabis, now 37, was a child prodigy: as a chess player he reached the rank of master in his early teens. Talk about talented.

He finished A-levels two years early and jumped straight into the world of work, landing a job at Bullfrog Productions, the legendary British game studio founded by Peter Molyneux responsible for hit games in the 80s and 90s including Populous, Magic Carpet and Dungeon Keeper.

His first job was designing levels on cyberpunk strategy game Syndicate in 1993, but he soon received a rapid promotion: he was made co-designer and lead programmer on the renowned Theme Park business sim, aged just 17.

The game was an international success, selling 3.5 million copies across 13 different platforms (including a Nintendo DS remake and PS3 port) and earned numerous thematic sequels and two direct sequels, including the BAFTA award winning Theme Park World. But Hassabis turned in fame and steadfast career at the studio to attend Cambridge, where he gained a double First in Computer Science.

© Bullfrog

After graduating, he returned to work on Molyneux’ latest title at the newly founded Lionhead Studios in Guildford (later purchased by Microsoft), Black & White, another god game with sophisticated AI mechanics working underneath, before breaking out to set up his own games company, Elixir Studios.

The studio’s first game, Republic: The Revolution, was even more ambitious than Black & White - you had to steer a grassroots political uprising in a fictional ex-Soviet country to power but was delayed several times and failed to match expectations.

Soon after that, Hassabis moved over to academia to pursue his interest in AI further, training as a cognitive neuroscientist. Even here, his interest in gaming continued. One of his experiments involved subjects playing a game similar to Pac-Man with much more at stake than usual: every time they were killed in the game, they were delivered an electric shock. At UCL in London, he received his PhD and made a monumental breakthrough in our understanding of episodic memory recall.

Then in 2012, he went on indefinite leave, striking out into the private sector once more, founding DeepMind with Shane Legg and Mustafa Suleyman, taking his discoveries about human learning and applying them to computers. The company operated in stealth mode for a long time - its website is still simply a landing page, though it reveals that DeepMind’s “first commercial applications are in simulations, e-commerce and games.”

© Elixr Studios

Google clearly saw value in its work and the staff they assembled, and acquired them. One of Re/Code’s sources claimed DeepMind used AI to help improve eCommerce recommendations, meaning Google could be better placed to compete against the likes of Amazon. Hassabis will now work alongside the likes of AI scientists including Jeff Dean and Anna Patterson, and celebrated author and futurist Ray Kurzweil, where he’ll likely be continuing his work on advancing machine learning. (We reached out to Hassabis after the acquisition was made public, but have yet to hear back.)

Of course, arguably he’s been doing that all along. His interest in complex AI is visible in the games he worked on even as a teenager. Theme Park for instance, was all about what visitors to your park (and your staff) made of it. The goal of the game, (Which came on five floppy disks at the time) was to create a profitable park that everyone enjoyed - and there were alot of AI tourists to please. Each of them expressed their own thoughts and preferences, grumbling if prices were too high, buying balloons if they were enjoying themselves, and throwing up on the path if your rollercoaster applied a bit too much G-Force.

The game inspired much of Bullfrog’s later work, including Dungeon Keeper and spin-off Theme Hospital, and arguably shaped the trajectory of Molyneux’ career for years to come - would he have been inspired to make Black and White, Fable or upcoming god game Godus without the success of Theme Park?

Few meanwhile have even tried to make as ambitious or as thoughtful a game as Revolution in the years since: why spend money simulating imaginary grassroots political movements when your game could have lovely predictable lasers and spaceships in it instead?

Perhaps with the Google purchase, Hassabis’ career has come full circle. Hassabis once said that gaming AI could help improve machine intelligence in general: "Games are good testing environments - they are not too simple, nor are they as complex as the real world."

Let’s hope real life doesn’t mirror gaming entirely however - alongside Republic, Hassabis also worked on a title called Evil Genius, a simulation in which your dastardly scientist-slash-despot has to take over the world. Fingers crossed, DeepMind’s valuable algorithms don’t morph into one of them.