AI has just made PAC-MAN
Last week marked PAC-MAN’s 40th anniversary since first taking the Japanese arcade market by storm. Ever since, the iconic maze game has been played billions of times. At its height during the 80s, approximately 250 million games of PAC-MAN were being played in the U.S. each week.
The global stardom of the greedy yellow ball has inspired NVIDIA to mark a new chapter in the game’s life.
Trained on 50,000 games – a powerful AI model called GameGAN has pulled off a feat that most video game programmers would scoff at. It’s generated a fully functional version of PAC-MAN. Without any game engine, or underlying code, or graphics design needed. The very same game that took Atari, at the time the largest video game maker in the world, some 17 months to develop – is now being procedurally generated by a computer. Wow.
GameGAN is the first of a new generation of neural networks that aims to replicate a game engine by harnessing GANs – generative adversarial networks. GANs are made up of 2 competing neural networks: the generator, and the discriminator. As an artificial player moves through the generated game, the GAN responds to the agent’s actions and procedurally generates unique game layouts. They’re able to create content that’s convincing enough to fool even the game designers themselves.
The reverberations will certainly be felt throughout the video-game industry. As Koichiro Tsutsumi from Bandai Namco puts it, “This research presents exciting possibilities to help game developers accelerate the creative process of developing new level layouts, characters, and even games!”
Video games are great, and I speak from immense first-hand experience. But it’s not just about games. Sanja Fidler, director of NVIDIA’s Toronto research lab, remarks that “We could eventually have an AI that can learn to mimic the rules of driving, the laws of physics, just by watching videos and seeing agents take actions in an environment. GameGAN is the first step toward that.”
Synthetic images produced by StyleGAN, another GAN framework created by NVIDIA researchers.
Image credit: Nvidia
Thumbnail Rights: Nvidia / Bandai-Namco