Techniques used to teach AIs to play video games may also be useful for teaching them to do more general tasks.
(CCM) — An artificial intelligence created by Microsoft researchers is now the world's greatest Ms. Pac-Man player, achieving the maximum possible score of 999,990 in the 1980s arcade game. The researchers, from a Canadian deep learning startup called Maluuba which was acquired by Microsoft earlier this year, used a branch of AI called reinforcement learning to teach the AI to master the game.
Ms. Pac-Man is held by many AI researchers to be one of the hardest video games to play well, but the AI mastered it by dividing the task into small pieces which were given to more than 150 different AI agents. Some concentrated on eating pellets in the game, while others were responsible for staying out of the way of ghosts. Then the researchers used a master agent which took suggestions from all the other agents and used that information to decide where to move Ms. Pac-Man.
This is similar to some theories of how the brain works, and it could have broad implications for teaching AIs to do complex tasks with limited information.
Image: © Microsoft.