A journal published in Nature on Wednesday, Feb. 16, highlights how researchers among two different groups came together to devise a newfound neural network AI strategy for Google's DeepMind in the process of controlling nuclear fusion.
Researchers amongst the Swiss Plasma Center of École Polytechnique Fédérale de Lausanne and DeepMind AI have been developing a new way for fusion reactions, specifically plasma configuration, to be made simpler via sophisticated autonomous AI control.
DeepMind's own neural network, which basically is the computer's process of mirroring the human brain, was first fed a simulation of the fusion reaction's main process. The AI learned how to alter the configuration of the 19 magnetic coils within the variable-configuration tokamak (TCV), initially allowing it to notice the changing shapes of the plasma.
Once DeepMind was able to process the alterations and learned how to change the settings, it was then ordered to recreate the varied shapes in the plasma, thus allowing it literal control over the complex process. The configuration shapes within the experiment that DeepMind recreated were of a snowflake, which is made to assist in even distribution of heat dissipation around the inner vessel of the fusion reactor, as well as a D-shaped cross-section that ITER (International Thermonuclear Experimental Reactor) will predominantly use upon its completion in France.
Martin Riedmiller, a control team lead under DeepMind, tells Wired, "AI, and specifically reinforcement learning, is particularly well suited to the complex problems presented by controlling plasma in a tokamak." The study and publication itself hope to brandish more leeway, avenues, and technology for the purposes of simplified fusion reaction processes, which may hopefully bring about a whole new form of renewable energy.
Related Article: DeepMind Ai 'AlphaFold' Gets Closer to Perfect Prediction of Protein Folding with Groundbreaking Accuracy
The tokmak of a nuclear fusion reactor is essentially the core of the whole chaotic process, wherein hydrogen atoms are flung at each other at dizzying speeds to create literal reactions through plasma and thus devise a new, more stable energy source from the chaos.
Via varied shapes and configurations, scientists can better understand numerous reactions that may or may not elicit different types of energy output through the tokmak's utilization of powerful magnetic coils.
However, the issue arises here that such complex and intricate workloads aren't exactly rocket science - or, more clearly, it's exactly that. It requires intense know-how and design and engineering input that aren't exactly commonalities.
Director of the Swiss Plasma Center, Ambrogio Fasoli, explains it best: "We need to be able to heat this matter up and hold it together for long enough for us to take energy out of it." This is where artificial intelligence can step in to aid in literally controlling this immense chaos at the center of the tokamak so scientists can have a better glance into the overall reaction.
Both research teams' gambit with DeepMind has seemingly paid off. The final result ended with DeepMind AI controlling and operating the magnetic coils perfectly in both the simulated trial run and the real-world process.
Even with the assistance from DeepMind, however, nuclear fusion wasn't exactly a simple process for the AI to learn. DeepMind AI researchers have dubbed it an "under-observed system," due to the fact that nuclear fusion is inherently complex with an unending list of variables. The plasma itself is constantly evolving when in the chamber, and it can't be measured at a consistent rate, either.
"This was a really big step forward for our algorithm because we could show that this is doable," explains DeepMind research scientist Jonas Buchli. "And we think this is definitely a very, very complex problem to be solved. It is a different kind of complexity than what you have in games."
It's a huge leap forward for physicists as more and larger nuclear fusion reactors are born going well into the future. The aforementioned ITER, which should see completion by 2025, will be on stage as the world's largest experimental fusion reactor. It's only the first of its kind in a long line of forthcoming nuclear fusion complexes that will reshape how we can experience renewable energy.
