Artificial intelligence in nuclear power plant construction should not be limited to document generation, chatbots, or stand-alone forecasting tools. A more consequential use is to connect the technical, commercial, and management processes of the project within a common machine-readable structure, from engineering and procurement through construction and commissioning.
My recent work, Recommendations for Implementing Artificial Intelligence Technologies in Nuclear Power Plant Construction Processes, explores this approach using two complementary models developed in my earlier research.
The first is a formal NPP life-cycle model built around five interconnected components: structure, semantics, document and data sources, quantitative attributes, and a meta-level responsible for orchestration. Its purpose is to create a unified and verifiable representation of the plant and project that can be processed by AI and machine-learning systems.
The second is a probabilistic optimization and forecasting model. Instead of treating an NPP project as a single deterministic schedule, it represents construction as a space of possible execution trajectories. Activity durations, alternative paths, risks, and project decisions can therefore be analyzed probabilistically, while optimization can account simultaneously for schedule, cost, risk, and other project objectives.
Together, these models provide a foundation for applying AI across the construction project rather than to isolated tasks.
The report identifies 18 application areas, including construction-readiness assessment, supply-chain forecasting, configuration and requirements management, design-documentation planning, licensing, critical-path analysis, risk and scenario modeling, cost and investment management, localization and import substitution, contract strategy, room and work-front readiness, commissioning, lessons learned, and project audit.
AI technologies such as LLMs, graph neural networks, conventional machine learning, RAG, analytical tools, and autonomous agents can then operate on top of this structured project environment.
The central idea is simple: AI becomes considerably more useful when it understands not only individual documents or datasets, but also how project objects, requirements, decisions, activities, risks, costs, contracts, and evidence are connected.
For complex projects such as nuclear power plant construction, the long-term objective should therefore not be a collection of independent AI applications. It should be an integrated project model that continuously accumulates knowledge, updates forecasts as new information becomes available, evaluates alternative scenarios, and supports decisions from the earliest project stages through construction and commissioning.
In this sense, AI is not a separate layer added to project management. It can become part of the project management system itself.