How to train your robots: Elon Musk's new AI gym

There's a new high-tech gym in the works, but it has nothing to do with toning muscles. 

Elon Musk, the tech-loving chief executive of both Tesla and SpaceX, is building an open-source "gym" for computer programmers to train their robots, Sharon Gaudin reported for ComputerWorld. Developers can share their findings on artificial intelligence (AI) systems as they work out.

"Nothing beats a competitive environment to motivate developers," said Patrick Moorhead, an analyst at Moor Insights & Strategy, to ComputerWorld. "It's like a monster truck rally for AI programmers."

Championing AI marks part of a calculated strategy by Mr. Musk, who has said AI could represent humanity's “biggest existential threat” because it could prove destructive in the wrong hands or simply go astray on its own.

Not one to wait idly for either inspiration or disaster, the billionaire behind PayPal and SolarCity has built this "OpenAI Gym" so innovators can train their artificially intelligent systems in the right path.  

"The irony is that Musk needs much-improved AI to make his businesses automated," Moorhead told ComputerWorld. "He doesn't hate AI. He just thinks it should be controlled before it gets too powerful and the machines turn on humans."

The gym is an early product of a $1 billion, non-profit research venture OpenAI that Musk, with others, launched in December, wrote The Christian Science Monitor's Max Lewontin:

Debates on the future of artificial intelligence often boil down to questions about whether the technology could help humans — detecting patterns that could help solve crimes or driving autonomous car, for example — or become the stuff of dystopian nightmares that have long fueled science fiction....

The new company will make its patents and research open to the public in a bid to increase transparency about AI’s potential rather than focusing on its commercial implications, say its backers — who include LinkedIn co-founder Reid Hoffman, venture capitalist Peter Thiel, the start-up incubator Y-Combinator, and Amazon Web Services.

OpenAI designed the gym, which is actually a series of labs for testing new technology, to develop algorithms for the non-profit's own research. The gym's exercises range from robot simulations to Atari games and are designed to develop reinforcement learning, the type of computer skills needed for motor control, and decision-making.

"Long-term, we want this curation to be a community effort rather than something owned by us," Greg Brockman and John Schulman wrote in an OpenAI blog post. "We'll necessarily have to figure out the details over time, and we'd would love your help in doing so."

Until now, non-standardized test simulations and progress reports have proven major obstacles in researching reinforcement learning. By opening the gym to the robot-savvy public, OpenAI hopes to build scientifically rigorous testing environments and thereby speed up development.

Once the gym is complete – programmers are still translating the environments from Python into other programming languages – developers can begin "exercising."

OpenAI advises developers to start with simple, small-scale tasks such as "toy text" and work their way into board games and robots in 2D and 3D. Each task is numbered so analysts can compare results directly. 

The nonprofit made their announcement on Twitter: