Inside of a tokamak – a donut-shaped vessel designed to accommodate nuclear fusion – exhibits a special kind of chaos. Hydrogen atoms slam into each other at unpredictably high temperatures, creating a swirling jet of plasma that is hotter than the surface of the sun. Finding clever ways to control and limit that plasma will be the key to unlocking the potential of nuclear fusion reaction, which has been promoted as the clean energy source of the future for decades. At this point, the fundamental fusion of science seems plausible, so what remains is an engineering challenge. “We need to be able to heat this matter and hold it together long enough for us to get the energy out,” said Ambrogio Fasoli, Director of the Swiss Plasma Center at the École Polytechnique Fédérale de Lausanne in Switzerland. from it.
That’s where DeepMind comes in. The artificial intelligence company, backed by Google’s parent company Alphabet, previously moved to video games and protein urgentand is working on a joint research project with the Swiss Plasma Center to develop an AI to control nuclear fusion.
In stars, also powered by fusion, the absolute gravitational mass is enough to pull the hydrogen atoms together and overcome their opposite charges. Instead, on Earth, scientists use coils of strong magnetic fields to limit nuclear fusion, push it into the desired position and shape it like a potter is in control. clay on wheels. The coils must be carefully controlled to prevent the plasma from touching the walls of the vessel: this can damage the vessel walls and slow down the fusion reaction. (There is very little risk of an explosion because fusion cannot exist without magnetism.)
But every time researchers want to change the configuration of the plasma and try different shapes that could yield more energy or cleaner plasma, it requires a huge amount of design and engineering work. Conventional systems are computer-controlled and based on careful models and simulations, but according to Fasoli, they are “complex and not necessarily always optimized.”
DeepMind has developed an AI that can control plasma autonomously. ONE paper published in a magazine nature describes how researchers from two groups taught a deep reinforcement learning system to control 19 coils from within the TCV, variable configuration tokamak at the Swiss Plasma Center, used to carry out the research. The study will inform the design of larger fusion reactors in the future. “AI, and reinforcement learning in particular, is particularly well-suited to the complex problems of manipulating plasma in tokamak,” said Martin Riedmiller, control group leader at DeepMind.
Neural networks – a type of AI setup designed to mimic the architecture of the human brain – are initially trained in a simulation. It started by observing how changing settings on each of the 19 coils affected the shape of the plasma inside the vessel. It is then made into different shapes to try to recreate in the plasma. These include a D-shaped cross-section close to what will be used inside the ITER (formerly the International Thermonuclear Experimental Reactor), a large-scale experimental tokamak under construction in Franceand the snowflake configuration can help dissipate the strong heat of the reaction around the flask more evenly.
DeepMind’s AI was able to automatically figure out how to create these shapes by manipulating the magnetic coils in the right way — both in simulations and when scientists performed similar experiments on real objects. fact inside TCV tokamak for simulation validation. Fasoli says it represents a “critical step,” one that could influence the design of future tokama or even speed up the path to viable fusion reactors. “It’s a very promising result,” said Yasmin Andrews, a fusion expert at Imperial College London who was not involved in the study. “It will be interesting to see if they can transfer this technology to a larger tokamak.”
Fusion presents a particular challenge to DeepMind scientists because the process is both complex and continuous. Unlike a turn-based game like Go, which the company has popular conquer with it AlphaGo AI, the state of the plasma is constantly changing. And to make things even harder, it cannot be measured continuously. It’s what AI researchers call an “under-observed system”.
“Sometimes algorithms that are good at these discrete problems struggle with such continuum problems,” said Jonas Buchli, a research scientist at DeepMind. “This is a really big step forward for our algorithm, because we can demonstrate that this is doable. And we think this is definitely a very, very complex problem that needs to be addressed. It’s a different kind of complexity from what you get in the game. “
