When Google’s AI is at the service of nuclear fusion

Nuclear fusion on the rise. This technique, which consists of combining atoms rather than separating them in two (nuclear fission) is more efficient, less dangerous, and produces less waste than our current reactors. Only problem, all this is for the moment only theoretical since we do not yet really know how to control it, despite rapid progress.

It’s not for lack of trying. Nations, start-ups and large companies, everyone is trying to apply what could become the energy source of the future. Among them, Google offers an original angle of attack: artificial intelligence.

The key to mastering nuclear fusion is to manage to control the plasma created by the fusion and to exploit the heat, more intense than that of the sun, that it releases. Google has therefore joined forces with the Swiss Plasma Center (SPC) of the Ecole Polytechnique de Lausanne to meet this engineering challenge.

DeepMind, Google’s AI, made a name for itself by successfully beating a go game master for the first time. It was then used in particular in the field of medicine. She is therefore now in charge of helping to develop a tokamak, the chamber where the fusion reaction takes place.

DeepMind to the rescue

The tokamak forms and maintains the plasma thanks to a powerful magnetic field. To prevent the plasma from coming into contact with the walls and damaging them, the magnetic coils must be precisely controlled, which requires “tedious calculations (…) necessary to determine the correct value of each variable of the control system”explains Federico Felici, an SPC scientist.

It is these calculations that DeepMind will be responsible for performing, in order to “to create and maintain a large set of advanced plasma shapes and configurations, including one in which two separate plasmas are simultaneously maintained in the chamber”.

The results of Google’s AI can then be tested in real conditions in the SPC reactor: by testing and finding solutions in a computer way, infinite time can be saved on experiments.

This collaboration, details Wired, could be particularly useful for the creation of larger tokamaks. “The more a tokamak is complex and has high performance, the more the control must be precise and of quality”says Dmitri Orlov, a scientist at the Center for Energy Research at the University of California, San Diego.

Leave a Comment