What Is DeepMind? The Artificial Intelligence Firm Bought By Google

on January 28 2014 9:53 AM
DeepMind Google Acquisition Goog AI Artificial Intelligence Robots
DeepMind, the company recently acquired by Google, uses deep learning techniques to program computers to learn from visual data, much like the human brain. DeepMind.com

Google (NASDAQ:GOOG) confirmed that it purchased DeepMind on Monday. The Mountain View, Calif.-based company reportedly paid upwards of $500 million for the artificial intelligence (AI) firm.

So what is Google getting for its half a billion? A company that's very good at making computers that think and act like humans.

DeepMind has not yet developed any commercial products. Its main asset appears to be its personnel, including dozens of experts in machine learning, a branch of AI that attempts to teach computers to think like humans. It’s best-known project was a computer system it taught to master Atari video games.

DeepMind says on its website that it combines “the best techniques from machine learning and systems neuroscience to build powerful general-purpose learning algorithms.” The London-based start-up says on its website that its first commercial applications involve simulations, e-commerce and gaming.

DeepMind co-founder Demis Hassabis, a neuroscientist and former child chess prodigy who spent 11 years in videogame design, including 2001’s “Black & White,” a “god game” that allowed players to choose between a path of good or evil, and 2004’s “Evil Genius.” In 2005, he began working in neuroscience and artificial intelligence, receiving a Ph.D. for the former from the University College of London in 2009.

Hassabis received renown and criticism for a 2007 paper that linked the mind’s process of imagination with memory formation. He founded DeepMind in 2011, along with Mustafa Suleyman and Shane Legg.

DeepMind’s Atari learning project consisted of a computer that was programmed to learn classic games through trial and error, by recognizing the visual data presented in the games. The computer system was able to beat expert-level human players at Pong and Breakout, but it was unable to beat humans at space invaders.

Re/code, which first broke the story on DeepMind, reports that the team will report to Google’s Jeff Dean, senior research fellow for “Knowledge,” the company’s term for Google search. Re/code says that DeepMind will, in fact, be working on Google search.

Google search is utilized for everything from research to shopping, and it is the backbone for Google Now, the service that powers voice searches on everything from the Moto X to Google Glass. By building AI that is capable of learning from its mistakes, DeepMind could impact further updates to the company’s search algorithm.

The "Deep" in DeepMind likely refers to how much of the firm's AI focuses on “deep learning” techniques, which allow computers to learn patterns from different forms of data without being specifically programmed to do so, including images. DeepMind could be used to help Google improve its Image and Shopping search functions.

The hand-eye coordination that DeepMind was able to instill in computer software for the purpose of Atari gaming could also come in handy with Google’s future robotics projects. Google went on a robot tech spending spree last year with the purchase of seven different robotics firms, with a potential focus on automated manufacturing.

Google bought Nest earlier this month, a company that connects traditional home appliances to the Internet for home automation -- including smoke detectors and thermostats. The company is also working on a self-driving car that uses AI along with GPS and Sonar technologies to navigate without a human driver. Facebook was also in negotiations to purchase DeepMind last year.

Both Re/code and The Information report that DeepMind asked Google to create an ethics board to govern how and why the tech giant would utilize its Artificial Intelligence.

Follow writer Thomas Halleck on Twitter @tommylikey 

More News from IBT MEDIA