Researchers from the Finnish Center for Artificial Intelligence FCAI and Aalto University have developed an AI system
Principles of operation of the manipulator. Image: I-Ju Chen et al., Nature Communications
Using their method, scientists managed to collecta star-shaped lattice of individual silver atoms in a very cold vacuum chamber. Forming a precisely defined structure is similar to moving balls across a Chinese checkers board, the scientists note, only in this case, instead of hands, tiny AI-controlled “tweezers” work to grab and drag each atom into place.
Traditionally, reinforcement learning is usedin video games or chess. It is a system where AI interacts with a real environment or model. In the course of learning, AI tries by trial and error to achieve the desired result, while for the right actions or results, it receives a reward. The new work is the first example of using reinforcement learning to manipulate particles at the nanoscale.
Construction of an atomic lattice. Image: I-Ju Chen et al., Nature Communications
The authors of the work adapted existing deep reinforcement learning for this purpose. It took the AI about one day to study the algorithm, and then about an hour to build the lattice of the desired shape.
The researchers note that precise movementatoms is difficult even for human experts. They believe that the introduction of machine learning can speed up the choice of parameters that are usually sorted out by people through trial and error. Together, this will help make a breakthrough in nanotechnology, scientists say.
Read more:
Hydrogen energy, material against cold weather and bioadditives against COVID-19: what scientists are creating in the North
The egg was dropped from space: look what happened to it
"The Walking Dead" existed millions of years ago: scientists told how they appeared