DeepMind AI Successfully Manages Superheated Plasma in Fusion Reactor
DeepMind collaborates with the Swiss Plasma Center to use reinforcement learning for controlling nuclear fusion plasma. This breakthrough demonstrates how artificial intelligence can solve complex physical challenges in the pursuit of clean energy.
DeepMind teams up with the Swiss Plasma Center at EPFL to apply deep reinforcement learning to a complex scientific challenge. The researchers train an artificial intelligence algorithm to control the superheated plasma inside a nuclear fusion reactor. This breakthrough, detailed in the journal Nature, shows how advanced AI systems tackle real-world physical problems.
Nuclear fusion requires extreme temperatures that turn matter into plasma, a state where atomic nuclei fuse to release massive amounts of clean energy. Because this roiling soup of particles is hotter than the center of the sun, scientists must contain it using powerful magnetic cages within a device called a tokamak. Controlling this plasma demands constant, precise adjustments to the magnetic field to prevent the plasma from touching and damaging the reactor walls.
The DeepMind team overcomes this hurdle by first training their reinforcement-learning algorithm inside a simulated environment. After mastering the simulation, the AI learns how to control and change the shape of the plasma, paving the way for more stable fusion reactions. Researchers believe this approach helps physicists better understand fusion dynamics and potentially accelerates the development of unlimited clean energy.