Google AI can now control magnetic fields in a fusion installation

A particular challenge in fusion systems is to keep the plasma stable in a magnetic field. This is a complex task due to the chaotic processes – but it must be solvable with a specially trained AI system. One is being worked on at DeepMind, the AI ​​company of the Alphabet/Google conglomerate. The AI ​​technology DeepMind is known for several Google projects that have produced quite impressive results.

For example, recognizing objects in images is different from controlling magnetic fields to controlling an extremely high-energy plasma. According to a report by American magazine Wired, the company is now working on the project with the Swiss Plasma Center. The joint research team has now published the results of the development work in a paper in the renowned journal Nature released. This describes how the AI ​​has been trained for its task of taking over an important part of the control of the TCV.

Hot Environment

This is a fusion device based on the tokamak concept. 19 magnetic coils generate the field in which the plasma must be located. These must be controlled in such a way that the extremely hot matter does not come into contact with the massive walls of the system, as no material is known to withstand such a hot plasma.

The AI ​​is trained by first learning how the magnetic fields actually affect the plasma. From then on, the AI ​​could train in simulations to get better and better at control – which is not easy, as the fields around the plasma are not rigid but are in constant motion. If the DeepMind system proves its worth at the Swiss factory, it could also be used in the much larger ITER fusion project in the future.