A Microsoft AI system has set the all-time high score for the game Ms. Pac-Man with 999,990 points – the highest possible score in the game. The AI mastered Ms. Pac-Man by assigning a number of AI agents different tasks and having them work collaboratively.
Add Ms. Pac-Man to the growing list of games being mastered by artificial intelligence (AI). An AI system from Microsoft set the all-time high score for Ms. Pac-Man with 999,990 points – the highest possible score in the game. The previous high score was 933,580 points set by Abdner Ashman, a human from New York.
Watch the YouTube video atop this page to see a small glimpse into how the AI played Ms. Pac-Man. Microsoft explains how a technique called Hybrid Reward Architecture used 150 agents for a specific task in the game, like finding pellets, or avoiding ghosts. This allowed Microsoft’s AI to play a perfect game.
“Then, the researchers created a top agent – sort of like a senior manager at a company – who took suggestions from all the agents and used them to decide where to move Ms. Pac-Man.
“The top agent took into account how many agents advocated for going in a certain direction, but it also looked at the intensity with which they wanted to make that move. For example, if 100 agents wanted to go right because that was the best path to their pellet, but three wanted to go left because there was a deadly ghost to the right, it would give more weight to the ones who had noticed the ghost and go left.”
Asilomar AI Principles: 23 Tips for Making AI Safe
Microsoft also says “the best results were achieved when each agent acted very egotistically – for example, focused only on the best way to get to its pellet – while the top agent decided how to use the information from each agent to make the best move for everyone.”
The AI played the Atari 2600 version of the game. The AI researchers say they chose Ms. Pac-Man as it was written to be far less predictable than the original version of the game.
The researchers say the AI technique that mastered Ms. Pac-Man could also be used to make advances in natural language processing. Other potential applications include helping a company’s sales organization make precise predictions about which potential customers to target at a particular time or on a particular day.