For years artificial intelligence learned by observing texts, images and videos on the Internet, but there is a problem, the real world is not made only of pixels, it has weight, friction, inertia and consequence, and it is exactly this gap that Google DeepMind is now trying to fill.

The company has just opened to the public a research prototype called Project Genie, powered by the Genie 3 model. On the surface it looks like a creative tool, you describe an environment or send an image and in seconds an interactive 3D world emerges that can be explored in real time, but beneath that there is something much more ambitious. Genie is not just creating scenarios, it is learning how worlds work, when you walk something moves, when you push the object reacts, when you change direction the environment responds coherently.

This means that AI is beginning to develop what the researchers themselves call intuitive physics, a basic understanding of cause and effect in space, and this advance solves one of the bottleneck problems of modern robotics. Robots need to learn millions of movements, but collecting data in the physical world is slow, expensive and risky. What DeepMind proposes is to reverse that logic, instead of teaching robots only in the real world, create practically infinite virtual worlds to train them.

Genie goes on to generate millions of simulated environments, within them, AI agents practice simple tasks, opening a door, avoiding obstacles, navigating through rooms, holding objects, falling, failing, it is a continuous cycle, worlds being created and agents training, models improving and better models creating even more realistic worlds, the ultimate goal is to train intelligences that will one day be transferred to physical bodies.

Companies like Electronic, for example, are already working on robots designed to receive this type of brain trained in simulation, today their virtual world still has limitations, the worlds last about a minute and there are delays in commands and physics sometimes behaves strangely, but the point is not perfection, it is direction, for the first time we see an AI not only learning about the world, but practicing within worlds that it creates itself.

And if the robberies are going to train among millions of virtual worlds, the brain that supports that needs absurd computational power and that is where artificial intelligence begins to point in other spaces.