The Race to AI-Powered Nuclear Energy: A Risky Venture or a Revolutionary Leap?
Power companies are turning to AI to accelerate the construction of nuclear power plants, but this move is sparking intense debates. Microsoft and Westinghouse Nuclear are leading the charge, aiming to revolutionize the energy sector. However, a report from AI Now warns that this could lead to catastrophic consequences, both in terms of nuclear safety and public trust.
The licensing process for nuclear plants is a lengthy and complex affair, designed to ensure public safety. But with AI in the mix, the game is changing. Microsoft's presentation reveals a plan to use generative AI to fast-track licensing, promising a reduction in time and cost. The company envisions training an LLM with existing licensing documents and nuclear site data, then using it to generate new documents, expediting the process.
But here's where it gets controversial: Heidy Khlaaf, AI Now's head scientist, argues that Microsoft's approach misses the point. Nuclear licensing is a meticulous process, not just a document-generation task. It involves understanding the safety of the plant, exploring design trade-offs, and communicating these intricacies to regulators. Khlaaf believes that AI, as currently proposed, won't support these critical objectives.
The Idaho National Laboratory and Lloyd's Register are already using Microsoft's AI for nuclear licensing, and Westinghouse is promoting its own AI system, Bertha, claiming to reduce licensing time from months to minutes. However, the AI Now report authors fear that this rush could bypass essential safety checks, potentially leading to disasters.
And this is the part most people miss: The use of AI in legal document writing has already faced challenges, with lawyers caught using AI-generated briefs in court, citing non-existent precedents and hallucinated cases. Could similar issues arise in nuclear licensing? Khlaaf warns that AI's minute mistakes, like software version control errors, could lead to misunderstandings and potentially catastrophic accidents, as seen in the Three Mile Island incident.
Moreover, the use of sensitive nuclear data to train AI models raises concerns about nuclear proliferation. Microsoft's request for real-time and project-specific data, according to Khlaaf, is a red flag, as building a nuclear plant requires many secrets not in the public domain.
Tech companies, including Microsoft, maintain cloud servers that comply with federal secrecy regulations and are sold to the US government. This raises questions about data security and the potential for nuclear secrets to fall into the wrong hands.
The Trump administration's actions have further fueled concerns. Bombing Iran over fears of a nuclear weapons program and selling weapon-grade plutonium to the private sector for nuclear reactors are just a few examples. Executive orders have also pushed for AI integration and NRC reform, raising doubts about the true motives behind these initiatives.
Matthew Wald, a nuclear energy analyst, offers a different perspective. He believes AI can be beneficial, citing the Three Mile Island accident as a case where AI-consolidated data could have prevented misunderstandings. Wald also highlights the industry's culture of safety, where engineers triple-check everything. However, he cautions against blind faith in AI.
Khlaaf and co-author Sofia Guerra, a nuclear safety expert, worry that framing nuclear power as a national security issue and using AI to speed up construction could backfire. They argue that nuclear energy's safety and public acceptance hinge on rigorous regulation and learning from past mistakes. By rushing the process and relying on unproven AI benefits, these experts fear that the risks of ionizing radiation and nuclear proliferation may be underestimated.
As the debate rages on, one thing is clear: the future of AI-powered nuclear energy is a delicate balance between innovation and safety. Will the benefits outweigh the risks? The answer may lie in the hands of policymakers, industry experts, and the public's willingness to engage in this critical conversation.
