INL: AI's Role in Nuclear Power

Idaho’s Joint Financial Appropriations and Finance committee concluded a three-day tour of Idaho Falls with a visit to Idaho National Laboratory Wednesday. The tour included stops at the College of Eastern Idaho, the Development Workshop, the Safety, Prevention and Resource Center, and INL.

A presentation on the laboratory’s concerns and applications of artificial intelligence sparked interest among JFAC members at INL. Chris Ritter, director of INL’s Digital Innovation Center of Excellence for scientific computing and AI, discussed the rapid changes in AI technology.

Ritter noted that AI services, like ChatGPT, initially produced inaccurate results when given prompts. He illustrated this with an image generated by AI several years ago depicting a woolly mammoth with five legs and missing a tusk. However, current AI services can now create accurate GIFs of mammoths.

“You can now add imagery and make things up that didn’t happen,” Ritter said. “From a national security perspective… we’re seeing those kinds of videos getting out there.”

Ritter and INL are concerned about the potential for both good and harmful uses of AI. “It’s incredibly important that America is the leader when it comes to AI,” Ritter said. He stated that the United States is currently behind China in AI technology, with China progressing in powering its AI facilities. “We need power,” Ritter said, emphasizing that nuclear energy is crucial for the country's AI leadership.

Officials expressed concern that China is developing infrastructure to support 32 gigawatts of power, while the US lags behind. Ritter, citing testimony from Eric Schmidt, former chairman of Google, noted the country will need 29 gigawatts by 2027 and 67 more gigawatts by 2030. “It’s roughly 100 gigawatts of more energy that we’re not producing,” Ritter said.

Ritter said that the US is attempting to catch up by funding more nuclear reactors to meet AI power demands. Meta, Amazon, Google and Microsoft are investing in infrastructure projects, but face challenges in securing sufficient fuel and materials for reactors. The INL aims to use AI to assess the suitability of materials for reactor use following irradiation tests.

Ritter explained that AI could automate repeatable testing processes, operating 24/7 compared to a worker's 10-hour workday. “We can get more fuel qualified more quickly, and get more use out of these federal funds we’ve already invested in these facilities,” Ritter said.

AI can also manage microreactors. Ritter stated that while a typical microreactor requires 16 operators, AI could reduce this number. “You have this kind of pairing of humans doing what we do good at…and then AI doing what it’s good at, keeping things in sync,” Ritter said.

Ritter mentioned two tests evaluating AI's ability to manage a reactor-like system on a smaller scale. In one test using a heat pipe, AI's performance was 0.3% different from a human operator. The other test, conducted with Idaho State University, involved connecting a digital five-watt reactor to the cloud and an AI model to predict student usage. “That was the world’s first digital twin limiting reactor (test) to happen here in the state,” Ritter said.