The researchers used a Hybrid Reward Architecture that utilized 150 AI agents that worked in parallel to master Ms. Pac-Man. Some AI systems were tasked with staying out of the way of ghosts, while others had the job of finding specific pellets. Suggestions from all of the individual AI agents were evaluated by a top agent, which used them to decide where to move the game character.
Doina Precup, an associate professor of computer science at McGill University in Montreal said that’s a significant achievement among AI researchers, who have been using various videogames to test their systems but have found Ms. Pac-Man among the most difficult to crack.More details at Microsoft.
But Precup said she was impressed not just with what the researchers achieved but with how they achieved it. To get the high score, the team divided the large problem of mastering Ms. Pac-Man into small pieces, which they then distributed among AI agents.
“This idea of having them work on different pieces to achieve a common goal is very interesting,” Precup said.
She said that’s similar to some theories of how the brain works, and it could have broad implications for teaching AIs to do complex tasks with limited information.
“That would be really, really exciting because it’s another step toward more general intelligence,” she said.