Inc.
Scientists at Lawrence Livermore Lab have built an AI model that predicts nuclear fusion experiments more accurately than supercomputers, saving time and money. While fusion energy is still elusive, this tool could speed up the race to achieve clean, limitless power.
A New AI Model Could Help Scientists Speed Up the Quest for Nuclear Fusion
Author: Lila Shroff
When it comes to alternative power sources, nuclear fusion is still a dream scenario. While using nuclear fusion to create commercial-scale energy has yet to be achieved, scientists say it could represent a tremendous transformation in both providing widespread energy while creating virtually no greenhouse gases or carbon emissions.
While the timeline for this breakthrough is still uncertain, scientists at Lawrence Livermore National Laboratory (LLNL) have developed a new artificial intelligence model that is smarter than all other previous prediction tools. And while it’s not a breakthrough in the fusion field, exactly, it could speed up the quest to create this more efficient type of energy.
The research, detailed in the journal Science, outperformed previous methods of predicting success in experiments. Older methods used supercomputers, but the AI is able to cover more parameters, which enhances its accuracy. The new tool, for instance, was able to accurately predict the results of a fusion experiment that was undertaken in 2022 at LLNL’s National Ignition Facility. In that experiment, it assigned a 74 percent probability for ignition. (Supercomputers only went as high as 70 percent.)
So what does this mean for future efforts to achieve fusion? The AI model, its backers say, will be better able to let scientists know if they’re on a potential right track. That could save both money and fuel, something that’s increasingly important as funding is far from certain at many labs these days.
For instance, in the experiment that was used to test the AI, scientists directed 192 powerful laser beams at a capsule filled with deuterium and tritium, managing to trigger a chain reaction under extreme pressure and heat. The cost of those experiments is high and running them takes time. At present, the National Ignition Facility, one of the leading labs working on fusion experiments, can only do a couple dozen per year. The AI guidance could make those more impactful moving forward.
Existing supercomputer prediction models take several days to examine the simulation code. To improve upon this, LLNL merged past data, expert knowledge, and physics simulations into a large language model. The new AI also replicates real-world problems, from instrument flaws to environmental factors. This gave researchers more insight into things that can go wrong, which helps them pre-emptively decide whether to proceed with an upcoming experiment.
Nuclear power today comes from fission reactions, which split atoms, creating heat that eventually boils water to turn turbines with steam. But the complex process also results in nuclear waste, which is long-lasting and environmentally hazardous. The upside is that they don’t produce greenhouse gases, while burning coal or natural gas does.
Fusion, meanwhile, generates electricity by combining atoms (much like the sun produces energy). That releases energy without emitting significant greenhouse gases or creating nuclear waste.
There are many private efforts on the hunt for a solution to the hunt for fusion. OpenAI’s Sam Altman has backed Helion Energy, which has started construction on a planned nuclear fusion power plant. The company is still working on how to achieve fusion, however.
Marathon Fusion, meanwhile, says its experiments to create fusion have let it stumble upon a method to transform mercury into gold. And Knoxville, Tennessee-based Type One Energy has teamed up with the Tennessee Valley Authority to explore plans for developing a commercial nuclear fusion plant.
The LLNL has seen the most success, though. In the 2022 experiment that was the one used to test the AI, it achieved results where the amount of energy created by the reactor exceeded the amount of energy needed to make it work. That, to date, has been the biggest breakthrough in the fusion field.
Credits: TCA, LLC.