Last month, you may have seen a group of researchers create a machine learning system that can change Grand Theft Auto V Become something It turns out that at about the same time, another group of AI enthusiasts were working on something more impressive, including Rockstar’s open world champion.On Friday, YouTuber Harrison Kinsley shared a video Show off GAN and steal the car, A neural network can generate a playable Grand Theft Auto VOwn game world.
Kinsley and collaborator Daniel Kukieła used GameGAN to make GAN Theft Auto, which was recreated last year Play the game by watching another AI. GameGAN, as the name suggests, is a generative confrontation network. Each GAN consists of two competing neural networks: a generator and a discriminator. The generator is trained on a sample data set and then told to generate content based on what it sees. At the same time, the discriminator compares the output of the generator with the original data set, and in the process “trains” its counterpart to the output content that is getting closer and closer to the source material.
“Every pixel you see here is generated by a neural network when I play a game,” Kingsley said in the video. “Neural network is the whole game. There are no rules written by us or others [RAGE] engine. “
Training GAN is a very GPU-intensive task. NVIDIA lent Kinsley a DGX A100 data center computer to make the project a reality.The system worth US$200,000 is equipped with the company’s eight And two 64-core AMD server CPUs. Kinsley and Kukieła use all this computing power to run 12 rule-based AIs simultaneously. These programs will drive on the same highway and collect the data needed by the neural network to start generating their own game world. The two also developed a super-sampling AI to clean the output of the neural network so that it does not look so pixelated.
As you can see from the video, the Internet has modeled a surprising number of systems in the game. When the car moves, the shadow below it and the reflection of the sun on the rear windshield also move. The mountains in the distance are getting closer. When Kingsley and Kojila first started training artificial intelligence, this was not something Kingsley necessarily expected artificial intelligence to do.
The gameplay also has a fantastic quality, thanks in part to the fact that neural networks cannot replicate all aspects of the game. Grand Theft Auto perfect. On the one hand, collisions will cause trouble for it. Kingsley said that once he saw an oncoming police car split in two and was about to hit his car.
If you want to try GAN Theft Auto yourself, Kinsley and Kukieła have uploaded the project to GitHub. They said that most computers should be able to run demos.
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 purchase goods through one of these links, we may earn member commissions.