New ‘gym’ for AI developers packed with Asteroids, Pong and Space Invaders

29 Apr 201616 Shares

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+Pin on PinterestShare on RedditEmail this to someone

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+Pin on PinterestShare on RedditEmail this to someone

Elon Musk’s $1bn OpenAI non-profit has just opened up its AI learning ‘gym’ to the public, with Atari games the perfect foil for a developing piece of software, it seems.

Released on Wednesday, the OpenAI Gym is a toolkit for developers to work on their AI programs, using reinforcement learning (RL) algorithms to garner improvements. The $1bn creation of a group including Peter Thiel, Elon Musk and Sam Altman, the gym was originally set up to help AI development itself.

Essentially, with all the vast sums of money these people have at their disposal, the research team, led by one of Google’s top machine-learning researchers, Ilya Sutskever, will get free rein to research what they feel is important, rather than what they can achieve under a limited budget, a researcher’s dream, without question.

But now the team wants to release it into the wild. The platform includes ‘environments’, which take in classic control and toy text, board games – Go, obviously – 2D and 3D robots, as well as Atari games. So Asteroids, Space Invaders, Pong and Video Pinball, along with 55 other classic games, are now free for your AI to learn against.

Asteroids

Atari’s legendary Asteroids game, via YouTube

“Over time, we plan to greatly expand this collection of environments. Contributions from the community are more than welcome,” said the organisation.

“Each environment has a version number (such as Hopper-v0). If we need to change an environment, we’ll bump the version number, defining an entirely new task. This ensures that results on a particular environment are always comparable.”

The reason OpenAI thinks RL is the way to go is that it’s a key subdivision of machine learning, concerned with decision making and motor control. It studies how an agent can learn how to achieve goals in a complex, uncertain environment.

“During the public beta, we’re looking for feedback on how to make this into an even better tool for research. If you’d like to help, you can try your hand at improving the state-of-the-art on each environment, reproducing other people’s results, or even implementing your own environments.”

The use of board games is less surprising, as it is here that AI has achieved it’s most famous successes.

At the end of the 1990s, AI had caught up with and overtaken humankind in the game of chess, with IBM’s DeepBlue first falling short against legendary player Gary Kasparov in 1996, before improving and beating him in 1997.

16 years later (back in March), AI had mastered an infinitely more complicated game called Go, with Google’s AlphaGo stunning world No 1 Lee Sedol with an emphatic 4-1 victory in the ancient game.

There are even projects looking at how AI robots can master the game of foosball.

The OpenAI gym is currently in open beta for researchers to start submitting their algorithms. It is initially available within Python with plans to update for any language and expand the selection of environments in the near future.

Space Invaders image via Reinis Ivanovs/Flickr

66

DAYS

4

HOURS

26

MINUTES

Get your early bird tickets now!

Gordon Hunt is a journalist at Siliconrepublic.com

editorial@siliconrepublic.com