A regulator's viewpoint on the use of Al in the nuclear industry

摘要

The regulatory position within the UK places the responsibility for safety on a nuclear site with the licensee. The licensee must produce a safety case that demonstrates that the plant is safe. Assessment of this safety case is performed by the regulator in the Health and Safety Executive (HSE) using HSE's published Safety Assessment Principles. This Paper presents the personal views of a regulator on the use of artificial intelligence (Al) in the nuclear industry. The Al-based systems considered are restricted to expert systems, neural networks and fuzzy logic. Some of the safety issues are discussed. The particular concerns associated with software-based systems are compounded by some of the inherent problems of Al-based systems. Such is the nature of these concerns that an acceptable safety case will be difficult to mount for safety systems and post-accident monitoring systems. However, given the present maturity of the technology, acceptable safety cases might be formulated in the areas of operator advisory systems designed to reduce accident frequencies, and robotics where dose reduction is important. There are also arguments that suggest that suitable safety cases could be formulated for Al-based plant control systems. Nevertheless, Al-based systems should be treated with caution — further research is required into suitable designs and safety demonstration techniques.