Several scientific articles point to the importance of education in the future generation of intelligent machines
We all have in mind the examples of robots that appeared on the big screen and that showed us what the future could be with these devices helping people. To become a reality, many scientists believe that artificial intelligence must provide a a step forward y join forces with human beings integrate into physical bodies.
Another challenge of robotics and artificial intelligence
We know how to make robots capable of different types of tasks and movements. In addition, every day we encounter examples of how artificial intelligence allows us to interact with the world in a more practical and simple way. Get integrate artificial intelligence into the physical world looks like logical step on this path, which is why science is already beginning to look into it.
For example, an article published in the scientific journal Science Robotics highlights what is called deep learning reinforcement. Use 50 cm robot tall, called OP3, theoretical computer scientist Tuomas Haarnoja tried not only to learn to walk, but also he learned to play footballa much more complex activity.
Google DeepMind scientist, co-author of the study, talks about two phases in teaching certain skills to the robot. First, is learning AIthrough reinforcement learning, k Get up now out of the country another AI to score goals. All available data, both from the robot itself and from the environment, are then used to generate the positions of the robot’s joints.
in second placeone train the new AIwho will be in charge imitate the first two and that they must manage to score goals against opponents who turn out to be identical versions. Joonho Lee, a robotics expert at ETH Zurich, said: “The football paper is incredible, given that we have never seen this level of resilience in humanoids.”
But what would happen if we extrapolated this example to robots that actually look like a human being? That’s what they did in research published in the journal Science Robotics. Ilija Radosavovic, a theoretical computer scientist from the University of California at Berkeley, used the so-called neural networks for train Digit the robotbelonging to Agility Robotics.
However, team of researchers used the type red neuron specialhe called transformer, similar to that used in language models such as ChatGPT. It was intended translate action observations into movements learned by observation.
After doing this work in the virtual world, robot I had to learn to function in reality. The researchers tested him with handballs to make him lose his balance, or with drawing steps, even ones he had never seen before. Pulkit Agrawal, an expert at MIT, talks to ScienceNews about the above research:
These items are equal to or have exceeded the manually defined drivers. It’s a turning point. With the power of data, it will be possible to unlock many more features in a relatively short time.
Join the Urban Tecno channel on WhatsApp
You can follow Urban Techno on Facebook, WhatsApp, Twitter or check out our Telegram channel to stay up to date with the latest tech news.
