Google has just unveiled Genie, a new artificial intelligence capable of generating video games. Currently limited to very simple platform games, it could create entire virtual worlds in the future.
Artificial intelligence is no longer limited to mastering games created by humans, it is now capable of creating its own games. She already knew how to generate amazing quality videos with text, images, music and OpenAI’s Sora for a week. Now, with Google’s new AI Genie DeepMind, the AI platform is taking on games.
Just provide an image GoogleGoogle Genie, and it generates platform games. The AI works with other AI-generated images just like real-world sketches or photos.
AI trained on public videos
The AI was trained on videos collected from the Internet, thanks to many of them specifically streamers on YoutubeYoutube Or TwitchTwitch. This means that there is no indication of which keys were used in the video, and so the AI had to infer the players’ actions from the image alone. Furthermore, the researchers specify that the model is not limited to platform games, and should be generalizable to any type of virtual world.
However, Google Genie is still in its early stages. This is a proof of concept, the first attempt to show that it is possible. It was trained on videos with a definition of 160 x 90 pixels, and therefore produces games with the same quality. However, given the way AI has made meteoric progress in other fields, complete games have been generated. 4K4K Maybe not too far now…
