40 years ago Pac-Man built an insatiable hunger for dots. Remarkably he’s always hungry to this day and Pac-Man has been indulged by millions of fans ever since. Pac-Man playing has seen lots of variations — some nice, others very poor. The original edition stays unchanged today.
In view of this, Nvidia ‘s AI was able to replicate a realistic recreation of the game from zero after seeing 50,000 episodes of Pac-Man being played.
On the 40th anniversary of Pac-Man ‘s iconic title, Nvidia, renowned for its graphic processing unit technologies, revealed that it has developed an AI to build an effective title clone simply by seeing it play 50,000 times — no human input required.
The artificial intelligence used to do this is named NVIDIA GameGAN, a neural network model that mimics a computer game engine by harnessing generative adversarial networks, or GANs. These GANS utilize a pair of opposing neural networks that function together to generate new material that looks impressively like the actual deal it was programmed to replicate.
And so, GameGAN recreated Pac-Man without anybody programming it with code without a game engine.
The company used the same GAN technology to create two further apps in 2019: GauGAN and GANimal. While the former could take user doodles and turn them into photorealistic scenes, the latter might actually place the appearance of one object — like your cat, for example — on another’s forehead.
As for GameGAN, Nvidia notes that this software the eventually be used by game developers to “automatically create models for new game stages,” or by AI experts to create “autonomous computer training simulator systems” in an efficient way.
Pac-Man ‘s AI-generated prototype will be open to the public in later this year.