Researchers at Maluuba, a deep-learning startup recently acquired by Microsoft, have developed an artificial intelligence (AI) system that became the first player ever to achieve a perfect 9,99,990 score in Pac-Man—an arcade video game.
The feat is notable, because when Pac-Man was developed in the 1980’s, it was programmed to be less predictable than the original Pac-Man, so that it would be tougher for players to beat it.
As opposed to developing a single intelligent agent to learn the game, the team used a number of simpler intelligent agents to each learn a single aspect of the game.
According to the researchers, by breaking up a problem in this way, it becomes easier to learn, as there are now many agents that learn very simple tasks, instead of just one agent that learns a very complex task. Pac-Man has remained elusive for years, owing to the game’s intentional lack of predictability.
Several people have tried to reach Pac-Man's top score, only coming as close as 2,66,330 on the "Atari 2600" version. The researchers believe that breaking up complex problems into simpler, smaller problems can make it much easier for deep learning systems to handle.
That, in turn, could be applied to lots of real-world tasks in the near future.
(At The Quint, we question everything. Play an active role in shaping our journalism by becoming a member today.)