The GAN Theft Auto project shows that AI is quite capable of creating a game environment that is actually playable, if not pretty.
GAN Theft Auto There is a project using Nvidia’s technology GameGAN (Generative Adversarial Network) using a . done to make KI to train who then tries to make his own version Grand Theft Auto V to manufacture. The result does not seem to throw the interested observer off the chair, but in any case is very clearly identifiable as GTA.
AI can already make simple games like pacman
It is also playable. The GameGAN fork shows how capable the technology is at handling more complex games than Pacman. Nvidia had a cult game from the eighties to mark its fortieth birthday can be recreated by AI.
Nvidia had stepped forward as the AI human developer. He tried to understand the rules of the game by looking at it and then tried to simulate it programmatically. Then he trained with 50,000 Pacman runs. Finally there was a fully functional version of the game which does not require any built-in game engine.
So KI recreated GTA5
Github users felt the same way centdex Front. In the video below, he shows that the AI was trained on various game scenes, showing it examples of cars in GTA5, including but not limited to separating the environment from the vehicle.
In the end, Centdex succeeded in creating a playable version of a – admittedly very simplified – section of the game. Visually, the result is not necessarily a pleasure. The atmosphere looks like you’re viewing it from eleven diopters, where you don’t need glasses.
Still, for the fact that the result was developed by AI and is completely playable, we can clearly be impressed. Coincidentally, it was already a few months ago An Intel AI with GTA5 deals.