Saturday, June 17, 2006

Dog neurons in a dish control video game (Quake 3). Skynet unavailable for comment.

The video links are spooky yet cool. Robot military, no draft.

"When the brain is first hooked up to the game, it doesn't know what it is doing so it randomly moves around," Thompson said. "After enough data flows to the brain it begins to change the neural network, it starts to control the character in a progressively more intelligent way."

Though the brain can successfully compete with a bot, the goal is to gain better understanding of how brains function, Thompson said.

-30-

Videos of the brain controlling the game are available:

Video 1: early stages of learning about 4 hours of learning

Video 2: slightly advanced, about 13 hour of learning

Video 3: more advanced, notice the collision detection and change of direction, about 42 hours of learning

Video 4: most advanced with combat, over 9 days of learning

UT SCIENTIST: DOG NEURONS IN A DISH CONTROL GAME

No comments: