If you’re interested in artificial intelligence (A.I.) and machine learning, OpenAI has a fascinating contest for you: design an A.I. agent capable of playing custom levels of the classic game “Sonic the Hedgehog.”
In case you’re a little fuzzy on your 1990s video games, “Sonic” is an extremely fast side-scrolling game in which you have to guide the title character (a blue hedgehog) through levels filled with all sorts of traps, falls, and enemies. That speed makes the game difficult for a human being—much less a software program—to fully master.
After you train or script your A.I. agent to play the game, OpenAI wants you to submit it to them via Docker container. The group will then evaluate the agent using “secret test levels,” and post a score on a leaderboard. Yes, software that’s 25+ years old is going to help prepare the software of the future.
“We believe that the next step for reinforcement learning is to leverage past experience to quickly learn new environments,” OpenAI wrote in the post announcing the contest. “Current algorithms are very prone to memorization and can’t adapt well to new situations. While this contest focuses on video game levels, we hope the winning techniques will be applicable to a wide variety of domains.”
The researchers with the best agent will receive a trophy of some sort once the contest finishes on June 5. More details (and a whole lot of code snippets) are available on OpenAI’s dedicated page.
OpenAI’s mission is to create “safe” artificial intelligence. To that end, it regularly releases research papers and open-source software tools for refining A.I. platforms. Much of its work focuses on training A.I. to play games, which offer the best of both worlds when it comes to improving artificial intelligence: a bit of real-world unpredictability embedded within a framework of hard rules.
While OpenAI’s founders are deeply concerned about the implications of A.I. set loose on the world, developers are far more sanguine about the technology. In a recent Stack Overflow survey, around 40.8 percent of developers said that “increasing automation of jobs” was an “exciting” aspect of A.I.; another 23.5 percent were enthused by the prospect of algorithms making important decisions. Compare that to the 19.8 percent of developers who thought that automation was “dangerous,” and it’s clear that the industry is more excited than scared about this future.