To celebrate Pac-Man’s 40th anniversary of the game first being released in Japans arcades and since going on a path to global stardom, Nvidia set out on the mission to recreate it the game purely with artificial intelligence and machine learning.

What is actually happening?

Nvidia used their own developed neural network model called GameCan that was designed to mimic a computer game engine. This model was trained through 100’s of hours of gameplay by analyzing how the game is interacted with, how the game looks, and what are the rules of the game.

You may be quick to think that the GameCan has recreated this game with traditional code that is normally associated with game development. But GameCan achieves the recreation through understanding what the game should look like, based on a users input.

A simplified example of this being, the GameCan model has watched Pac-Man for long enough to understand every time Pac-Man moves over a small yellow circle, that circle gets removed from the screen. So the model now understands visually what that looks like when Pac-Man eats a dot.

The model is not outputting a traditional coded game, it’s actually just outputting frame by frame what the model thinks the game should look like for every frame based on all the rules and visual elements it has been learning over the 100’s of hours.

We wanted to see whether the AI could learn the rules of an environment just by looking at the screenplay of an agent moving through the game. And it did.

#ai #bootcamp #machine-learning #flatiron-school #nvidia