DeepMind Pools Data Learning From Robot Labs

General AI, or an AI model that can train itself, is talked about as an inevitable outcome based on the current scaling of technology.

However, most experts agree that an AI that can think independently is still years away.

One of the challenges is that most artificial intelligence on the market today, including even the large language models such as ChatGPT, are focused on a task or a small set of capabilities.

If you look at science fiction as a predictor, the AI models that we are playing with today will ultimately inhabit a robotic body like C3PO.

Robots are in wide use today but similarly to the AI models, they are typically only good at a handful of tasks which they can do over and over again with a great deal of precision.

Many of the advances in the world of robotics are coming from academia, out of the labs at universities around the world.

But this knowledge is siloed to a particular lab and while universities are much more willing to share advances than corporations, it is difficult or impossible for researchers to get a clear picture of all these advances in one place.

That is where a new initiative by Google’s machine learning company DeepMind came to the rescue.

DeepMind announced recently that it is funding an open-source library for robotic labs to share called Open X-Embodiment.

The initiative seeks to provide a growing library of robotic tasks that can be joined together to create the science fiction version of a robot that is capable of doing many things well.

But to reach a general level of artificial intelligence requires that the model teach itself which remains in the world of science fiction and theory.