Playing a Neural Network’s version of GTA V: GAN Theft Auto

Wednesday 30th June, 2021 - Bruce Sterling

*This isn’t too surprising if you’ve been watching AI master games like chess and go, and this works on similar principles.  But it’s still startling to see an AI simulate a computer game without using any of the computer game’s software.


*Instead, it just has a huge database of how the game visually behaves on a screen, and it can generate play action in real time using a neural net as a “simulator.”  Except that game simulators simulate the code of other games.  This one doesn’t even bother with the code — it just simulates the way the game appears through a massive statistical study of the behavior of its pixels.

*I wish I could explain this better, but I don’t think the proper language for this has been invented yet.


by sentdex

GAN Theft Auto is a Generative Adversarial Network that recreates the Grand Theft Auto 5 environment. It is created using a GameGAN fork based on NVIDIA’s GameGAN research.

With GAN Theft Auto, the neural network *is* the environment and you can play within it.


Unboxing and reviewing the DGX Station A100 80GB:…

Neural Networks from Scratch book:

Channel membership:…
Support the content:…