Computer artificial intelligence is still a mystery to many of us. While we can teach a computer to pick up a can, solve logic puzzles, or even complete some advanced games, we still haven’t reached a level of sophistication that the fictional worlds of Star Wars and Star Trek clearly think we should.
Still, it is still fun to see the small steps forward we make in all areas of AI, because these will probably directly affect videogames someday in someway. Even cooler yet is when videogames are used for AI research–like Ms. Pac-Man.
In a nutshell, AI researchers have taught a computer how to not only play Ms. Pac-Man, but also score better than a human opponent would, over an average of fifty games. This was done with a technique known as reinforcement learning, a type of testing which has been used on a variety of other games like chess and Baldur’s Gate.
Taking this along with rule-based policies and a cross-entropy method, the computer eventually was able to play Ms. Pac-Man at a very high level. This was impressive due to the ghost paths not being fixed, as they are in the original Mr. Pac-Man.
Interestingly enough, the computer did exhibit the human tendencies to lure ghosts into corners or keep track of when the time window to eat ghosts was about come to a close. How this will inflence the future of videogames really isn’t quite clear yet, but suffice it to say, any progress is good progress.
[Via RobotWorldNews]
