The Artificial Intelligence DeepMind has already beaten the best players of Go, the weather, or even the 3D structure of proteins. Now, it is battling fusion energy. And it seems to be doing quite well. The fusion AI controls the tokamak and shapes the extremely hot plasma to stay where it needs to stay.
Image by Dimitris Christou from Pixabay
- Be also sure to check out my other posts and follow me @kralizec and subscribe to my Youtube channel at Kralizec Gaming Youtube Channel
The British DeepMind is a subsidiary of the Alphabet Inc. Conglomerate that was a few years birthed by Google. Recently, they achieved significant successes in the field of artificial intelligence when the AI DeepMind beat the best players of the game Go, successfully predicted rain, or showed the years-old problem with finding the 3D structure of specific proteins. But that’s not enough.
DeepMind recently gave its impressive know-how in AI to the tamers of fusion plasma from the Swiss Plasma Center (SPC) which is part of the Ecole Polytechnique Federale de Lausanne (EPFL). And it seems to have been a good thing.
The fusion scientists at SPC use a classical tokamak. This very expensive doughnut in which extremely powerful magnetic fields keep plasma and astronomically high temperatures stable. This plasma is where fusion reactions should take place one-day atoms should fuse instead of spliting while releasing gigantic amounts of energy.
SPC has an experimental tokamak called TCV (variable-condition tokamak) in which it is possible to experiment with plasma under various conditions and various configurations. The experts from SPC are using it to find new ways to control and regulate the plasma so it doesn’t cause problems.
Their fusion simulator is based on more than 20 years of fusion research and is being constantly updated. Yet, it is still slowed down by demanding calculations that need to be performed for every variable in the control system of the tokamak. This was the place where DeepMind could shine.
DeepMind developed a new AI that was trained at the fusion simulator at SPC. The AI learned to find effective strategies for controlling the tokamak leading to desired plasma configurations. The trained AI was then given control of the real tokamak and it managed to create and control many different shapes of plasma. In one of the experiments, the AI even managed to hold two plasma masses at the same time.
The goal now is to breed a new generation of AIs to act as closed-loop controllers that will learn to function in environments with highly complex dynamics.
Sources:
- https://www.nature.com/articles/s41586-021-04301-9
- If you like the content I’m producing about science maybe you will like the content I produce about gaming as well! Be sure to check out my other posts!