Google’s DeepMind Artificial Intelligence Program Excels Further
Google probably never thought they would be bested by a Ms. Pac-Man arcade game, but nevertheless, this is currently the case. Last year, Google’s DeepMind system, an artificial intelligence software, made the news due to its ability to completely trounce any human opponent in the classic arcade game Pong. Since then, Google has continued to work at developing DeepMind’s abilities, including their arcade game prowess, but there continues to be one title that outsmarts the program: Ms. Pac-Man.
How Does DeepMind Work?
Google’s DeepMind is a “neural network” designed to replicate how the brain processes information, but based on mathematical calculations for reasoning power. The system is built in layers, with each layer responsible for feeding information to the next until reaching the top tier group of computer node neurons, which then makes decisions based on the processed information.
DeepMind has recently been revised to improve its efficiency. Researchers reduced the chance the system would expect a positive outcome in any given situation. The changes to the algorithm resulted in better results overall in problem-solving puzzles like classic arcade games.
Why is Ms. Pac-Man a Challenge to Beat?
Google’s DeepMind is not taught how to play classic arcade games. Instead, the game’s information is fed into the system pixel by pixel, which allows the system to determine specifics about the game including the distance between objects. The system must use trial and error to complete the game’s levels and improve their abilities over the course of a week.
Ms. Pac-Man is one of a few classic arcade games that DeepMind is unable to master, with its scores falling well below the average human score. Why?
Achieving success while playing Ms. Pac-Man is dependent on remembering patterns in the maze. DeepMind is only able to recall the last four frames of each game during play which translates to continued failure navigating the mazes in this classic game. So for now, humans rule the scoreboard when it comes to Ms. Pac-Man, but future revisions to the DeepMind algorithm addressing the system’s long-term vision and memory may give the artificial intelligence program the boost it needs to defeat human player’s scores at this game classic as well.