DeepMind unveils self-learning RoboCat

An anonymous monkey admirer once said, “Cats and computers both have one thing in common – they both rule the Internet.”

At Google’s DeepMind, researchers recently married artificial intelligence with a robot called RoboCat, and while it hasn’t yet ruled the Internet, it’s expected to leap into the future world of self-learning.

Using the same technology behind a large variety of languages, the DeepMind team, made up of more than 30 researchers, said it had achieved great success with a robotic cat that not only learned new tasks quickly but also could improve its performance by creating its own actions.

“RoboCat has good training,” said DeepMind in a paper published on the preprint server. arXiv. “New jobs that learn, are better at learning new jobs.”

Until now, robots have often performed specific, pre-programmed tasks. With the use of large languages, the robots’ abilities began to increase, although training on large amounts of information required a lot of time.

DeepMind said Robocat, however, can quickly learn new tasks, such as putting pieces of different shapes into the right holes or putting fruit in a bowl. It was then able to go further and perform more complex operations “using data from millions of trajectories” from previous operations with new, automated data.

“This change was due to the development of RoboCat’s knowledge, similar to how people develop different skills as they develop their training in a given environment,” the researchers said.

As RoboCat developed its skills, its new behaviors were transferred to other robots that also built those skills.

The robot changed its behavior in between 100 and 1,000 movements from the human-controlled arm. Spin-off models were trained on specific tasks and the data was fed into the training pool.

Although RoboCat initially achieved a 36% performance on untrained tasks, it improved its performance over time. Through self-education, his success doubled.

“RoboCat learns faster than other modern models,” said DeepMind researchers. “It can take a new job with less than 100 shows because it is based on large and diverse data.”

This development is seen as a major step forward in the development of robotics research, “since it reduces the need for human training, and is an important step in the creation of a ready-made robot.”

The paper, “RoboCat: A Self-Improving Foundation Agent for Robotic Manipulation,” was published June 20.

Will such robots eliminate the need for human intervention?

This question was answered 100 years ago in the 1921 play “RUR: Rossum’s Universal Robots,” a story written by Czech writer Karel Čapek.

The game imagined a factory that produced artificial humanoids that worked continuously and eventually reduced labor costs by 80%. The word “robot” was first used in the play, after the Czech word “robota,” which means “forced labor and serfs.”

In the end, the robots rebelled and extinguished the humans.

RoboCats, we can hope, will be friendly.

We should also remember what comedian Will Rogers once said: “It’s easier to let the cat out of the bag than to put it back in.”