NVIDIA’s GameGAN Shocks the World With An AI-Generated Version of PAC-MAN

On May 22, the iconic video game series, Pac-Man turns 40. And to mark this special occasion, a team of researchers from NVIDIA has come up with something that will blow your mind – an AI-generated version of the game.

Get… set… go, GameGAN!

NVIDIA‘s researchers had already been working on developing a powerful machine learning tool, which is also the first neural network that will be able to copy and reproduce a game engine. This tool in question is GameGAN.

Wondering how it will be able to copy and reproduce the game engines? Well, it’ll be done using Generative Adversarial Networks (GANs).

Hold on: things are about to get a bit technical here.

A GAN is made up of two competing networks: one acts as a generator and the other as a discriminator. Both networks have a shared goal of analyzing a source carefully to create a reproduced version of it after gathering enough information through this observation.

nvidia game gan pacman

While this version isn’t an exact carbon copy, it can still be passed off as real.

Keep in mind that there isn’t any underlying engine and instead, it’s the AI that produces a new frame for every on-screen based event.

A quick look at GameGAN’s journey

Interestingly, the GameGAN Model had 50,000 training sessions to observe, study, and learn every aspect of the original game. It was only after this meticulous process was the recreation possible.

Again, we’d like to emphasize that any input of the original game’s engine or boundaries was completely absent.

NVIDIA’s vice president of simulation technology, Rev Lebaredian told Mashable, “It learned the rules of Pac-Man. It observed it just like a human might.”

He sounds proud and certainly should be.

The only problem? The AI model doesn’t recreate audio. So, whatever you hear was added over GameGAN’s playtime.

Exciting Times Ahead

Exploring broader applications of this technology is on the cards. Not only will it make the lives of game developers easier, but it might also be useful for increasing the efficiency of autonomous robots.

While the possibilities are endless, we sure need to keep an eye out for what these GANs can bring to the table in the next couple of years.

About The Author