Microsoft AI has achieved the highest score possible in Ms. Pac-Man Video Game.
Maluuba, a Microsoft owned firm has achieved the highest score of 999,990 before it rolled over to 0. This is 4 times higher than any human score.
The 1980s Atari version of the game was used, but most pacman games are similar in nature.
Hybrid Reward Architecture, a method of reinforcement learning was used to achieve the score. 150 AI “agents” collaborated to beat the game. Some agents focused on eating pellets and some agents avoided ghosts. A top agent, or “manager,” got feedback from the others and made decisions on which direction to move.
Harm Van Seijen, a research manager with AI firm Maluuba, said the best results happened when an agent acted egotistically.
He also stated that “A lot of companies working on AI use games to build intelligent algorithms because there’s a lot of human-like intelligence capabilities that you need to beat games.”
Google’s DeepMind is commonly accepted as the leader on AI research, after it beat Go.
Creators of the arcade version of the game said that it had been designed to be simple but impossible to beat, so people can put more money into the machines.
Ms.PacMan is an unpredictable game. Beating this game is a milestone in deep learning. Researchers can help AI make decisions in other real-world situations.
Boeing’s testing of automated flight next year will benefit from this type of application.