Microsoft AI Plays Perfect to Beat Ms Pac-Man!

.

A Microsoft AI has achieved the dream most high score chasers can only dream of. The AI has successfully played a perfect game of Ms Pac-Man by scoring 999,990. Microsoft has said that the achievement may lead to new develops in AI software.

Didn’t this happen last month?

No. You might be forgiven for confusing this with the Google-developed AI which beat the world champion at ‘Go’. This is, however, a totally separate entity.

In a report by the BBC, the AI by Microsoft was specifically designed to learn and develop to the point of scoring a perfect game at Ms Pac-man.

For those of you unfamiliar with Pac-Man (really?) the game involves a maze, collecting pellets while avoiding 4 ghosts. The ghosts in themselves have unique AI depending on their colour. The AI by Microsoft, therefore, had to learn and adapt to the patterns to score the perfect game, something Humans have yet to achieve.

When it comes to Pac-Man, the best a human has ever managed is a ‘kill screen’. That, however, is achieved simply by running out of levels. The perfect score has yet to be achieved by humans.

What was the point of this?

The AI was developed with the hope to help AI assist in complex tasks to help us mere humans.

It is not all sunshine, I’m afraid. Prof Nello Cristianini, a computer scientist from the University of Bristol has given the following ominous warning: “It is exciting that so much progress is happening today in AI, however, we should remember that historically AI has not always been able to replicate results in games when transferring methods to real world problems. This should be kept in mind whether we talk about Jeopardy, Chess, Go or Ms. Pac-Man.”

Harm Van Seijen, a research manager with Maluuba who ran the tests has said “There’s this nice interplay between how they have to, on the one hand, co-operate based on the preferences of all the agents, but at the same time each agent cares only about one particular problem.”

The AI is therefore quite interesting. In a sense that it doesn’t solely take the most immediate logical move to hand. It considers over 100 independent ‘thoughts’ on the subject and selectively analyses which are the most advantageous. Such processing could help us to develop better choices in technologies moving forward.

There does seem to be a bit of a trend for picking older retro video games. There are the occasional new games thrown in, GTA 5 was decent used for traffic analysis. I do, however, eagerly await to hear that Donkey Kong assisted in deciphering the Voynich manuscript, that missile command was used to control Trident and that Centipede might be used to repel an alien attack. You never know!