Artificial intelligence software designed by Microsoft Corp. has achieved something that was previously never thought possible: a perfect score in the computer game Ms. Pac-Man.

Using computing power and programming power on what would appear to be a frivolous use of AI. But Microsoft claimed in a blog post today that the method used could have broad implications for teaching AI agents to do complex tasks that augment human capabilities.

The AI platform for the achievement came from Maluuba Inc., a deep learning company Microsoft acquired in January. Programmers for the platform used what they call Hybrid Reward Architecture, which uses a “divide-and-conquer” computing method. The method uses an algorithm design paradigm based on multibranched recursion that recursively breaks down a problem into two or more sub-problems of the same or related type until these become simple enough to be solved directly.

The solutions to the sub-problems are then combined to create the solution to the main problem, in this case how to get a perfect score on the Atari 2600 version of Ms. Pac-Man. The team working on the AI used 150 agents that were each focused and rewarded for a specific completed task in the game, such as finding pellets or avoiding ghosts, allowing the AI to play a perfect game.

“This idea of having them work on different pieces to achieve a common goal is very interesting,” Doina Precup, an associate professor of computer science at McGill University in Montreal said on the official Microsoft blog. Precup claimed that the method is similar to some theories of how the brain works, meaning that 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 added.

For those gaming purists out there who want to know what a perfect score in Ms. Pac-Man is, the answer is 999,990. The previous record was 933,580 points.

Photo: fleur-design/Flickr