Google's AI DeepMind Trumps Humans at Atari 2600 Games

By Peter R - 26 Feb '15 07:23AM
Close

An artificial intelligence system developed by Google has outperformed the best human players at 29 video games of the 80s, without any human assistance.

The 29 games are from Atari 2600, a legend in video game parlance, which the AI called DeepMind not only learned but mastered, beating the best human players by a big margin, Tech Crunch reports. DeepMind relies on deep neural networks and reinforcement learning system that rewards the programme for success. This meant DeepMind did not require human intervention once it started gaming. At the start it was clumsy but as it learnt it got better, just like humans.

According to The New Yorker, DeepMind was acquired by Google in 2014 for $ 650 million. Ever since, it has beaten humans at 29 of the 49 games it learnt including Fishing Derby, Freeway and Kung-Fu Master. The New Yorker further pointed out that DeepMind's abilities are far more complex than what IBM's Deep Blue could do to beat Chess grandmaster Gary Kasparov.

DeepMind's creator Demis Hassabis feels that the programme's success is an indication it could become capable of performing real world tasks in the distant future. Hypothetically speaking, success at gaming means the system can use its abilities in problem-solving and thus by extension, real world applications. However, the immediate goal for the DeepMind team is to train it to excel at 90s games, including Doom.

Fun Stuff

Join the Conversation

The Next Read

Real Time Analytics