SRP webinar: Teaching AI the Laws of Physics – Paradigm Shift in Radiation Protection

As Artificial Intelligence (AI) becomes increasingly integrated into the physical world, its application within nuclear safety–critical domains has come under heightened scrutiny. While AI has demonstrated considerable effectiveness in perception-based tasks—such as processing data, images, patterns, and speech—its deployment in nuclear safety applications presents significant challenges. In particular, the data-driven and often opaque “black box” nature of conventional AI systems limits their ability to provide transparent, auditable rationales for their outputs. This lack of explainability raises regulatory and industry concerns, as it may introduce unknown or unacceptable safety risks if AI-generated recommendations are implemented without sufficient understanding or validation.

To address these challenges, an emerging field of research is focused on the development of Explainable and Physics-Informed Artificial Intelligence. In this approach, AI models are explicitly constrained and guided by the Laws of Physics, system dynamics, and governing equations embedded directly within the neural network architecture. This paradigm supports the development of trustworthy AI systems that are capable of delivering accurate, rapid, robust, and interpretable solutions while remaining consistent with established physical principles.

This presentation will describe the methodology and challenges associated with training AI systems to recognise, respect, and enforce physical laws when formulating predictions and decisions. It will demonstrate how Physics-Informed AI can transform radiation protection by ensuring that AI outputs remain physically plausible and scientifically defensible, thereby improving accuracy, reliability, and interpretability in complex and safety-critical environments. Finally, emerging applications across nuclear and environmental safety, as well as medical physics, will be discussed, highlighting the potential of this approach to support regulatory acceptance and operational deployment.