DeepMind, Google’s AI division, recently unveiled Genie 2, an AI capable of creating interactive 3D worlds in less than a minute. Although the result is impressive, it raises, as is often the case with generative AI, many ethical questions regarding the methods used to achieve such a result.
Genie 2: when AI creates video games
DeepMind introduced Genie 2an innovative AI model capable of generating rich, interactive 3D worlds from simple text descriptions and images. Designed as a foundation for creating virtual environments, Genie 2 uses big data to simulate realistic interactions, sophisticated animations and dynamic behaviors of characters and objects.
This model, successor to Genie, is capable of transforming an image and a textual description into interactive scenes in real time. These worlds enable actions like jumping or swimming, through keyboard-mouse interactions, and include animations, object behaviors, and non-player characters (NPCs). This is clearly creating a mini-game in seconds. Environment generation can last up to a minute and offer different points of view, such as first or third person.
Genie 2 is a world model, which means it can simulate virtual worlds, including the consequences of any action (e.g. jumping, swimming, etc.). It was trained on a large-scale video dataset and, like other generative models, demonstrates various emergent capabilities at large scale, such as object interactions, complex character animation, physics, and ability to model and therefore predict the behavior of other agents.
Trained on videos, Genie 2 is able to create universes from a simple image, which raises questions about the collection of data necessary to develop it and give it this exceptional creative capacity. Although this technology promises unprecedented immersion, its legal framework, particularly with regard to copyright, remains to be clarified. This applies to all companies with large language model (LLM)-based generative AI.
First review of Genie 2
The examples presented on the official page are truly breathtaking. Whether it is the modeling of water, the smoke or even the interactions with NPCs, each element is consistent. Without forgetting the diversity of worlds created with realistic 3D effects. The renderings are still quite blurry in some cases, but that does not detract from the technological prowess.
As of now, Genie 2 is not capable of generating worlds longer than a minute. However, with the rapid evolution of generative AI, it is plausible that within months or even years, Google could create full 3D mini-games simply by writing a line of text to an AI. The video game world should really start to worry.
