Microsoft created an artificial intelligence-based computer system that just shattered the record for the highest score in “Ms. Pac-Man” by achieving the highest score possible.
A team of researchers from Maluuba — an AI research company recently acquired by Microsoft — used a special type of AI called “reinforcement learning” to play an emulation of Atari’s “Ms. Pac-Man.” The AI system was able to achieve a score of 999,990 before it reset back to zero.
According to Highscore.com — a popular resource for tracking and collecting high scores in video games — the previous high score made on the Atari 2600 “Ms. Pac-Man” emulation was 266,360, by a player in Brazil.
The approach used to master “Ms. Pac-Man” is called Hybrid Reward Architecture, which uses a series of agents that have a designated function such as finding pellets, avoiding ghosts, eating the blue ones and going after the large bouncing fruit. The AI system prioritizes all of the variables in real time to make sure Ms. Pac-Man gets a perfect score.
You may be wondering why they chose to use the 1980s video game “Ms. Pac-Man.”
According to Steve Golson, co-creator of the arcade version of “Ms. Pac-Man,” Microsoft needed to use a game that was easy to play in theory, yet nearly impossible to master.
Rahul Mehrotra, a program manager at Maluuba, believes video games like “Ms. Pac-Man” are complex because of many situations one can face while playing.
“A lot of companies working on AI use games to build intelligent algorithms because there’s a lot of humanlike intelligence capabilities that you need to beat the games,” Mehrotra said in a Microsoft press release.
Check out the video of the Maluuba AI system achieving the high score. The video also shows the game resetting back to zero once it reaches the high score.