Last month, you may have seen a group of researchers create a machine learning system that could transform the presentation of Grand Theft Auto V in something that . It turns out, at about the same time, another group of AI enthusiasts were working on something even more impressive with the Rockstar open world title. On Friday, YouTuber Harrison Kinsley shared a video showing GAN Theft Auto, a neural network that can generate a playable stretch of Grand Theft Auto VThe game world only.
Kinsley and co-star Daniel Kukieła have done GAN Theft Auto with GameGAN, which last year recreated watching another AI play through the game. GameGAN, as its name suggests, is a generative adversary network. Each GAN consists of two competing neural networks: a generator and a discriminator. The generator is formed on a sample of data and then told to produce content based on what it saw. Meanwhile, the discriminator will compare the generator output with the original dataset, in the process of “coaching” its counterpart to the output content that is always closer to the source material.
“Every pixel you see here is generated by a neural network while playing,” Kinsley says in the video. “The neural network is the whole game. There are no rules written here by us or by u [RAGE] engine. “
Forming a GAN is a very GPU intensive task. NVIDIA has lent Kinsley a DGX A100 data center computer to make the project a reality. The $ 200,000 system comes with eight from the company and two 64-core AMD server CPUs. Kinsley and Kukieła used all that computing power to manage 12 rule-based AIs simultaneously. Those programs would drive the same stretch of highway, collecting the data the neural network needed to begin generating its game world. The two have also developed AI supersampling to clean the output of the neural network so it doesn’t look too pixelated.
As you can see from the video, the network models a surprising number of systems from the game. As the car moves, so does the shadow below and the reflection of the sun on the rear windshield. The mountains in the distance are approaching as well. That’s not something Kinsley necessarily expected AI to do when he and Kukieła started training AI.
There’s also a quality like a dream to the game and that’s partly due to the fact that the neural network doesn’t replicate every aspect of it. Megane V perfectly. For one, the collisions cause trouble. Kinsley says there was an example where he saw a police cruiser approaching split in two just as he was about to crash into his car.
If you want to try GAN Theft Auto for yourself, Kinsley and Kukieła have loaded the project in GitHub. They say that most computers should be able to run the demo.
All products recommended by Engadget are selected by our editorial team, independent of our parent company. Some of our stories include affiliate links. If you buy something through one of these links, you can earn an affiliate commission.