The nonprofit OpenAI, along with the support of Tesla and SpaceX founder Elon Musk, believes artificial intelligence and machine learning can advance a lot faster if people start programming A.I. to crack high scores in Atari 2600 games.
That’s because artificial intelligence programs, like the tech industry itself, suffer from a diversity problem. OpenAI says there are essentially two problems with programming A.I. right now: There’s not enough variety, and the language to communicate with other developers is not the same.
The OpenAI Gym beta, launched today, claims it can fix both of those problems. It’s a development toolkit for those in the A.I. community, and challenges coders to come up with solutions to myriad of problems, including programming a virtual cart to balance a pole, teach a two-bit car how to climb out of a valley and, yes, beat the high score in Atari’s Asteroids.
If a lot of people come together to develop Atari wizard programs, perhaps it will allow others to find weaknesses in a subsection of machine learning called reinforcement learning (RL), which has been the driving technology behind programs such as Alphabet’s Deep Mind Alpha Go program that beat a grandmaster in the ancient Chinese game of Go earlier this year.
