Artificial intelligence has entered the nuclear industry and its early benefits are just the tip of the iceberg

A collaboration between startup Atomic Canyon and Oak Ridge National Laboratory enabled the construction of a sentence embedding model using 53 million pages of Nuclear Regulatory Commission documents. Understanding nuclear terminology opens doors for artificial intelligence to revolutionize many processes within the industry. The nuclear power sector is considered by many onlookers to be a slow-moving industry. Strict regulatory frameworks mean that extensive testing, documentation and approval processes are necessary for any changes to nuclear facilities or procedures. Furthermore, nuclear power plants represent enormous investments, requiring owners and operators to be cautious when it comes to making changes to proven designs. Major upgrades must be thoroughly validated given the complexity of nuclear systems and the potential consequences of failures, and the studies conducted as part of the process can take years. The sector relies heavily on experienced staff, who are trained to follow detailed procedures, meaning they can be resistant to adopting new approaches. So in the end, things often change slowly. However, changes are happening and the technology has made its way into the nuclear industry as the development of advanced reactor designs and small modular reactors have accelerated innovation in this field. Recently, artificial intelligence (AI) even found its way into the nuclear conversation when a startup called Atomic Canyon partnered with the Department of Energy’s Oak Ridge National Laboratory (ORNL) to develop an advanced AI model capable of understand complex nuclear terminology.

AI training

“The first thing you need to build artificial intelligence is a data set – you need access to information,” said Trey Lauderdale, founder and CEO of Atomic Canyon. CURRENT. Lauderdale is not a nuclear expert, but he has founded and supported multiple companies over the past fifteen years with a focus on using technology to improve processes, which is what Atomic Canyon’s AI platform is designed for for nuclear power plants, manufacturers of next generations. generation reactors, and government and national laboratories. “One thing we quickly realized about nuclear energy is that there is a huge amount of data. The Nuclear Regulatory Commission – the NRC – actually has a database called ADAMS (which stands for Agencywide Documents Access and Management System), where there are all kinds of public information available for anyone to view, and it’s all available on their website. Lauderdale said. When the Atomic Canyon team started building AI models and experimenting with the ADAMS dataset, the experts quickly discovered a problem: all AI models that are publicly available got confused when they used “nuclear words.” encountered. Lauderdale explained: “The nuclear language is very complex. It contains all kinds of acronyms and words that these AI models haven’t seen enough examples of yet. What ends up happening is that the AI ​​hallucinates. That’s what AI stands for: ‘It makes things up.’ As you can imagine, making things up in an industry like nuclear energy is very, very bad.” The Lauderdale team realized that they did not necessarily need to create a new large language model (LLM) to solve the problem, but rather just build sentence embedding models for AI applications so that nuclear terminology could be understood. “To do that you need access to a lot of so-called GPUs: graphics processing units,” says Lauderdale. A typical startup could raise millions of dollars and buy a bunch of GPUs to do a project like this, but Atomic Canyon had a better option: partnering with the government. ORNL is home to Frontier (Figure 1), a supercomputer that was touted as the fastest in the world upon its debut in May 2022 and retained that title through the most recent May 2024 rankings. “It was quickly discovered that this was a project was that was worth of the world’s fastest supercomputer: the ability to train AI models on nuclear terminology and then get an output that is essentially a more advanced search application that can be used to find documents,” said Lauderdale.

The results were astonishing. Within just six months, the team developed an advanced AI model that could understand the complex nuclear terminology. This specialized open-source AI model has set new benchmarks for accuracy, efficiency, and speed in AI search. The model was developed as open source and is available to ORNL, the Nuclear National Laboratory Complex, independent researchers and nuclear institutions. It will also be integrated into Neutron, Atomic Canyon’s AI search platform. The open source aspect was important to Lauderdale. “I would argue that the ethos of the nuclear industry is inherently open source in nature,” he said. In many industries there is fierce competition and companies are reluctant to share information that could jeopardize their competitive advantage. But Lauderdale said the nuclear industry is the exact opposite. “The statement I’ve heard over and over is: ‘An accident in any factory is an accident in any factory.’ The result is that you share so much data – whether it’s with the NRC, whether it’s with INPO (Institute of Nuclear Power Operations), an organization that provides quality metrics to all these nuclear power plants – that there’s an ethos of openness, transparency and the back and forth sharing information again,” he explained. “By moving to the technology and the code that we built, we used government resources to build this. There were the NRC data. There was Oak Ridge. And we believe that enabling artificial intelligence to understand nuclear terminology is so fundamental to any AI application that’s going to be built that we want to make this code open source, which is a way of saying, we want to make it ensuring that every party, even our competitors (people who may be building competing applications) can use this toolset when building their own apps. And all we ask in return is that they withdraw from the project if they make improvements (by adding various features),” he said.

The AI ​​revolution

Still, Lauderdale suggested this is just the beginning. “Everyone is hearing that there is a major AI revolution underway, and we are all starting to realize as a society that the energy demands for artificial intelligence will be astronomical, to the point that Three Mile Island will be reopened to a Microsoft database. downtown,” Lauderdale said. “We’re at the tip of the iceberg, because all the hyperscalers are talking about 10x, 20x, crazy growth, and that requires reliable energy that’s available 24/7, that’s safe and ideally doesn’t emit carbon. So nuclear energy is that path.” However, the path is not necessarily simple or easy. “We believe that not only does AI need nuclear, but nuclear actually needs AI,” Lauderdale predicted. “I think there is an opportunity to apply artificial intelligence in very safe ways and in fundamental ways.” While Lauderdale doesn’t think AI is ready to run nuclear power plants on its own, he does believe it could benefit operators and developers. “The ability to have AI help people search and find documents is a very valuable thing,” he said. “It’s where we started. And from there, once you have that foundation, you build layer upon layer of more advanced applications.” “Where the first wins can really come is in the licensing process,” said Tom Evans, leader of the High-Performance Computing (HPC) Methods for Nuclear Applications group in the Nuclear Energy and Fuel Cycles Division at ORNL (Figure 2). CURRENT. “The knowledge-based barriers to understanding and navigating the licensing process are high, and if you can do something to significantly reduce those barriers, right out of the gate you have already made a huge, huge improvement in the entire process and to the prospects of actually being able to use nuclear energy more economically,” he said.

Still, Evans suggested there are multiple vectors through which AI can play a role. One area is in complex design analysis. Evans said ORNL often helps the NRC and other industry stakeholders by conducting these types of analyzes for them using supercomputer models. He explained that analysts usually start by searching through previous analyzes to find something similar that they can use as a reference. From there, they adjust the input to account for the differences in the new design and then run simulations. However, Evans noted that just identifying the most applicable past analysis to serve as a starting point can take a lot of time. This is where AI could improve the process. Instead of an analyst having to sift through vast amounts of data, the AI ​​tool, which has literally been trained on 53 million pages from NRC’s ADAMS database, can quickly produce the most relevant files. This would save the analyst a lot of trouble. Atomic Canyon also has bigger ambitions. Lauderdale said his team is in discussions with several companies in the nuclear field that have proprietary data sets into which they want to integrate AI. He said his group could install an enterprise version of the software that would process the company’s proprietary data, allowing users to search through their large amount of internal data. “That’s kind of the next version of where we’re going,” Lauderdale said. In addition to his own company’s ambitions, Lauderdale suggested that the nuclear industry is not as slow-moving as some people might think, and could in fact become a leader in the AI ​​revolution. “One of the key ingredients you need to build great, world-class AI models is data. And the more data you have, the better models you can build,” he said. “Because the nuclear space has documented so much information, I actually think this space has an opportunity to really be a thought leader and an innovator in artificial intelligence, and that’s what really excites me.”

—Aaron Larson is the editor-in-chief of POWER.