California’s only operating nuclear power plant, Diablo Canyon, announced plans last month to use artificial intelligence systems to keep it running, sparking questions regarding the use of AI in such advanced facilities.
In November, Diablo Canyon’s operator, Pacific Gas and Electric Company, said it would be partnering with California-based nuclear AI company Atomic Canyon to build AI systems, called Neutron Enterprise, within the nuclear plant. These systems are expected to be used to sort through millions of documents to create plans for the facility to meet licensing requirements and assist with regulatory planning in the coming years.
At the time of the announcement, PG&E even suggested the software could be used for additional tasks, such as maintenance scheduling.
Throughout 2024, technology giants such as Microsoft, Google, Meta, and Amazon increased their investments in nuclear energy to enhance and support their AI developments. Comparatively, very rarely has it been seen the other way around. However, that’s not to say it’s not being considered.
“You can imagine that it’s probably not super common, but it is an idea that has been around for a long time,” Katya Le Blanc, a human factors scientist with the Idaho National Laboratory, told the Washington Examiner.
TRUMP’S ENERGY AGENDA IN THE NEXT ADMINISTRATION: A CLOSER LOOK
Le Blanc said nuclear power plants store massive amounts of documentation in the form of procedures, drawings, regulatory documents, and more. With much of the cost associated with running and operating such a facility going toward managing those documents, she explained that the industry has been exploring different technologies to make it easier to access and sort through that documentation.
There are also other ways AI can be used in relation to nuclear facilities, such as modeling reactor core designs and optimizing designs.
Though no matter the current use, for Xingang Zhao, an assistant professor of nuclear engineering at the University of Tennessee’s Tickle College of Engineering, AI cannot be used without its human counterparts.
“AI is supposed to generate outcomes that can help humans, human operators, and domain experts to make decisions,” he said, pointing to the difference between decision-making and decision support.
“It is a decision-making ability that AI can enable, but I don’t want people to see AI as a replacement for humans. They should be used by humans to ensure control and decision-making by humans and meet the needs that the humans have,” Zhao continued, later adding that when it comes to nuclear reactors, the industry isn’t going to let AI “take control.”
As with introducing any new technology, using AI within the nuclear energy industry can bring some risks. As AI has continued to advance in recent years, so have the discussions on safety and security around the tech. At the same time, nuclear power has, for decades, required detailed attention to safety.
“If you combine these two, it’s really natural to think that there are risks,” Zhao said.
Though, as the technology has developed, many experts are in agreement that these risks are quite minimal, particularly with human moderation.
Yavuz Arik, a nuclear and energy consultant with the American Nuclear Society, told the Washington Examiner the industry can’t trust AI to reveal when it is processing wrong data and making inaccurate calculations. That is where human operators and engineers are still stepping up.
But, when the data are right, it can be a win for everyone, Arik explained.
“When an engineer sits in front of a computer, he has to think of each possibility and then go and calculate each possibility,” he said. “With AI, you can just let it run loose. And with supercomputers these days, the sky’s the limit.”
There is some agreement that eventually, this will lead to AI being used for operations within nuclear power facilities. Le Blanc described it as a “natural progression” within the nuclear and AI industries.
CLICK HERE TO READ MORE FROM THE WASHINGTON EXAMINER
“If you look at how AI is used in other industries for scheduling or these data-driven approaches, there aren’t — it’s not that hugest step,” she said, indicating that there should still be caution when taking that “natural step.”
“I do think you have to be conscious of how you’re doing it and how you’re using it because there are limitations in these technologies,” Le Blanc said. “Making sure that you understand exactly what your problem that you’re trying to solve is, and what the requirements are for that problem, and then applying the AI in a way that is consistent with the data that you have. … I think it’s a very, very good opportunity. But you need to do it very mindfully.”