GAN Theft Auto - GTA 5 made by artificial intelligence

That artificial intelligence is evolving daily is evidenced by the various filters, settings, and options on smartphones, cameras, and other devices we use regularly. But all this is slowly going a step further in terms of video games, where Nvidia's artificial intelligence called GameGAN (generative adversarial network) has demonstrated its power, making a playful section inspired by the GTA V game.

If you’re wondering what GAN is and on what principle it works, here’s a brief explanation: GAN is an artificial intelligence made by merging two neural networks, one of which is a generator in charge of learning from real-life examples such as images of people, to make "Fake content" that can pass as real. The second network is the discriminator, which is in charge of keeping the generator "honest", that is, separating the true from the false.

"GAN can make realistic images of human faces, even though the face doesn't belong to any real person," they say from Nvidia, and last year they discovered this technology with a demonstration of Pac-Man made without a single line of written code.

Now, this has been repeated by Youtuber Harrison Kinsley and his collaborator Daniel Kukiela with the help of GAN on the example of GTA V.

They trained GAN with a dozen simultaneous information on the highway scene in the game and with the help of that, artificial intelligence learned how the car moves, how it responds to controls, and the like. Of course, the simulation is not ideal, most of you will first notice the low-resolution graphics, and then the torture when interacting with other cars, but if you pay attention to the shadow under the car, the sun's reflections on the rear windshield, are the results to be respected, since the human hand did not write a single line of code. 

This is all an indication that it is only a matter of time before artificial intelligence will be able to make a realistic game without the help of the human factor, and until that moment comes, you can try GAN Theft Auto, as it is available on GitHub.