Machine Learns Games 'Like a Human'
Researchers at Britain's University of Leeds have developed a computer system that uses observation and mimicry, as humans do, to teach itself to play the children's game "scissors, paper, stone." The system, dubbed CogVis, constructs its own "hypotheses" about the rules of the game by studying video and audio input of human players for specific patterns. The system watched people playing the game with cards marked with scissors, a piece of paper, or a stone; the players were instructed to announce when they won or when the game ended in a draw. After several hours of observation, CogVis was able to successfully call the outcome of each game. CogVis team member Chris Needham explains that the system's visual processor deconstructs action into periods of movement and inaction, and then distills color- and texture-based features; the addition of audio allows the system to formulate theories about the game's rules via inductive logic programming. CogVis was demonstrated in December at an event sponsored by the British Computer Society, and won the prize for Progress Towards Machine Intelligence. "A system that can observe events in an unknown scenario, learn and participate just as a child would is almost the Holy Grail of AI," notes the University of Leeds' Derek Magee. Portsmouth University researcher Max Bramer thinks machines could one day use CogVis technology to learn to control maintenance robots or spot intruders on video, while Imperial College London AI expert Stephen Muggleton says enabling the system to learn more complicated games such as tic-tac-toe will be a major challenge.
Read More