Google's Genie AI: what can it do, what games can it create

Genie is an incredibly complex AI technology that serves the purpose of creating new interactive worlds. Here we explain in clear language what Genie is and what it is capable of now and what it promises to do in the near future.

What is Genie?

A month after the launch of the Gemini AI model, Google revealed Genie (short for Generative Interactive Environments), a demo of its new generative artificial intelligence model. Genie has been created in collaboration with the University of British Columbia as part of Google's DeepMind AI technology. Its breakthrough nature, according to Google, is that it can independently create games and other interactive visual products from any video, image or text.

How does Genie work? For example, to create a new game, this AI model just needs to watch and analyze the mechanics of videos from the Internet. Google calls its game creation engine a latent action model, which converts raw video frames into discrete tokens that are molded into a new game using dynamic design technology. Genie recognizes not only external actions in these videos, but also operates on hidden actions and parts of images and is able to model deformable objects.

Genie is able to analyze an impressive number of Internet videos - hundreds of thousands. However, for this AI model, one image, photo, text or even a simple sketch is enough to create a game dimension from them. "This opens the door to a variety of new ways to generate and step into virtual worlds, for instance, we can take a state-of-the-art text-to-image generation model and use it to produce starting frames that we can then bring to life with Genie," says Google.

What games can Genius generate?

So far, Genius can create online 2D side-scrolling, single-image platformers in the style of Super Mario Brothers. However, the company assures that soon the AI model will learn to generate 3D worlds and master the finest mechanics of movement.

According to the creators of DeepMind, the resulting game worlds are "crisp and aesthetically pleasing." It was no coincidence that Google chose the name of the wizard from the lamp as the name. "We use a dataset of around 200k hours of films from 2D platformers and train an 11B earth product… In an unsupervised way, Genie learns varied latent actions that handle figures in a dependable method," Google DeepMind developer Tim Rocktäschel wrote on X.

As the corporation hints, games are just the beginning. Plans for Genie involve technology to create "universal AI agents of the future." We can't wait to try out Google's new AI tool for ourselves, and hope that a public version will appear on the market sooner or later.