The nuclear energy industry should prioritize “safety and security” in its deployment of artificial intelligence systems, according to a recent report from the Nuclear Regulatory Commission (NRC).
The report from the NRC, in partnership with the United Kingdom’s Office for Nuclear Regulation and the Canadian Nuclear Safety Commission, provided potential requirements for AI use and governance that industry may adopt. The implementation of AI requires continuous monitoring, a modular approach, and data integrity alongside clear documentation, according to the regulatory agencies.
“The rapid pace of recent AI development is somewhat antithetical to the slow and methodical change process that the nuclear industry traditionally follows,” the report reads. “Nevertheless, the primary goal for the nuclear industry and regulators with respect to AI systems will be maintaining adequate safety and security while benefiting from their deployment.”
One of the largest concerns cited in the report is the unpredictability of AI system failure, complicating failure analyses and resulting in longer times to identify and resolve issues compared to traditional systems.
“Currently no method exists to quantify the failure probability of an AI component within a system, which makes it difficult to trust AI components to perform a function with any level of integrity,” the report reads, stating that “incorrect output” from AI systems may result in “no effect to unacceptable risks to people, the facility, or the environment.”
Solutions to this include improved system performance tracking, developing security strategies to counteract system vulnerabilities, and using iterative testing methods.
Recommended monitoring methods include tracking performance metrics including computational efficiency and model bias, anomaly detection and the implementation of alert systems, and utilizing modular monitoring to isolate issues.
While the report notes that the recommendations are not expected to be adopted on a consensus basis across countries, it says that “it is nevertheless important to examine some foundational tenets all parties should consider,” while adopting AI in the nuclear sector.
