Microsoft's AI Just Beat the All-Time High Score in 'Ms. Pac-Man'

Friday, 16 June 2017 - 11:59AM
Technology
Artificial Intelligence
Friday, 16 June 2017 - 11:59AM
Microsoft's AI Just Beat the All-Time High Score in 'Ms. Pac-Man'
Image credit: Alexisrael

When you think of the classic game Ms. Pacman, images of dimly lit arcades and long hours spent trying to beat the top score come to mind. It's a game that many, many kids (and adults) over the years have worked to beat. That is, until AI stepped into the picture.

Maluuba, an AI startup recently acquired by Microsoft, decided last month to see what would happen if you reached one million points in Ms. Pacman. Of course, the team at Maluuba isn't that inhumanly talented at the game—they had some help from their AI. While others (human and AI alike) have tried to reach the score before, Maluuba succeeded. But the moment wasn't as gratifying as they'd hoped: "It just reset to zero, it was kind of disappointing," said Rahul Mehrotra, a program manager at Maluuba who watched the maximum score, found to be 999,990, surpassed.

Image Credit: Maluuba / Microsoft

 

This was still a big win for the team, however. While the practical benefits of being able to totally beat Ms. Pacman might not weigh in too heavy, the software and algorithms that allowed the AI to evade ghosts, navigate the maze, and munch on pellets could be extremely useful for other tasks. Because the company is a part of Microsoft, they think that this technology could be applicable to sales, prioritization, and other exciting business applications.

While Atari games might seem like just fun ways for researchers to test their machine learning technology, they have proven useful time and time again (for example, testing the efficiency of deep learning by playing Breakout). And, while it might ignite some fury in those who have spent tireless hours and quarters trying to master arcade games, creating machines capable of beating these games quickly and systematically could seriously help to advance machine learning capabilities across the board.

Science
Weird Science
Technology
Artificial Intelligence
No