Tuesday, May 10, 2016 by Greg White
Elon Musk wants to make robots sweat. The CEO of SpaceX has created a $1 billion state of the art “Robot Gym” to train artificial intelligence (AI) systems. The intent of the gym, according to Musk, is to make robots as versatile as humans with respect to thinking.
Musk joined forces with Sam Altman to create the A.I. non-profit research company. The OpenAI Gym Beta made its debut on April 27th. The gym provides developers with the means necessary to improve their reinforcement learning (RL) programs, according to Nature World News.
RL is part of a study on AI, which centers on fulfilling various tasks rather than abiding by simple rules. Popular Science explains it functions by rewarding the AI whenever its algorithm is executed efficiently.
“It consists of a growing suite of environments (from simulated robots to Atari games), and a site for comparing and reproducing results. OpenAI Gym is compatible with algorithms written in any framework, such as TensorFlow and Theano,” Greg Brockman and John Schulman wrote in the company’s blog.
They believe new “environments” provide an opportunity to test AI capabilities. For example, the AI can be placed in an environment where it has to finish various tasks or play games. Since the open gym is funded by a non-profit organization, anyone can use it free of charge.
These environments include board games like Go, where the AI can interact on a 9×9 and 19×19 boards. It also encompasses 59 Atari games, including Pac-Man, Space Invaders and Asteroids, according to Tech Times.
OpenAI reported that public Gym Beta can be accessed and downloaded on their company website. The gym was originally designed as a means to test research projects within the company. By joining forces with Musk, however, the gym is now available to the public. “We hope it will be just as useful for the broader community,” OpenAI said.
To keep tabs on the accomplishments of the robots, a message board will list the top robo performers. These rankings are not based upon high scores. Rather, they are based upon the versatility of the AI system.
“What matters for research isn’t your score (it’s possible to over fit or hand-craft solutions to particular tasks), but instead the generality of your technique,” Brockman and Schulman noted.
They added the environments are written in Python right now, but the team plans to make them easier for any language to use.
OpenAI hopes the public will provide constructive feedback for the beta version of their AI gym in order to make necessary changes and improve their research tool.