Sign In

Communications of the ACM

ACM TechNews

AI Learns to Defy Laws of Physics to Win at Hide-and-Seek


View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
Artist' conception of bots playing hide-and-seek.

Researchers at the OpenAI artificial intelligence laboratory developed bots that trained themselves to cooperate by playing hide-and-seek.

Credit: OpenAI

Scientists at the OpenAI artificial intelligence (AI) laboratory have developed AI bots that trained themselves to cooperate by playing hide-and-seek.

The team had the bots play the game in a simulated environment containing fixed walls and movable boxes; each bot had its own perspective of its surroundings, and could not directly communicate with other bots.

The bots that hid quickly deduced the fastest way to fool seekers was to find objects in the environment with which to conceal themselves; the seekers learned they could manipulate objects like ramps to overcome obstacles like walls.

The bots learned that cooperation—like passing objects to each other or co-building a hideout—was the quickest way to win.

Said Chelsea Finn at Stanford University, “The main limitation of this kind of work is that it is in simulation.”

From New Scientist
View Full Article

 

Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account