Ever wonder what Grand Theft Auto V would look like if artificial intelligence were to generate the world in real-time? If so, then look no further than GAN Theft Auto, which basically uses an AI-powered neural network to create a fully playable section of the game. This was made possible by using GameGAN by NVIDIA, or a generative model that learns to visually imitate a desired game by ingesting screenplay and keyboard actions during training. Read more for a video and additional information.
GameGAN works by designing a memory module that builds an internal map of the environment, allowing for the agent to return to previously visited locations with high visual consistency. Plus, it’s caspable of disentangling static and dynamic components within an image making the behavior of the model more interpretable, and relevant for downstream tasks that require explicit reasoning over dynamic elements. In other words, it can swap different components of the game to build new ones that do not yet exist. More information here.
- NVIDIA GeForce RTX 3050 Ti 4GB GDDR6 up to 1585MHz at 60W (75W with Dynamic Boost 2.0)
- Intel Core i7-11370H processor (12M Cache, up to 4.8GHz)
- 15.6” 144Hz IPS-Type Full HD (1920x1080) display with adaptive sync
Every pixel you see here is generated from a neural network while I play. The neural network is the entire game. There are no rules written here by us or the [RAGE] engine,” said its creator.