A New View on Metrological Maintenance of AI-based Systems

摘要

The paper discusses the issues of information trustworthiness that are actual for Industry 4.0 and, to a greater extent, for Industry 5.0. Such issues are important regarding both weak and strong artificial intelligence (AI). Corresponding information includes measurement results as well as knowledge being introduced in the system during its training, e.g., working conditions, measurement ranges, ethical norms of robots. The authors support the analogy between the evolution of the nervous system of living beings and artificial intelligence development. This analogy gave them grounds to propose the training of technical systems and their maintenance to be carried out similarly to those in human upbringing, with prioritising certain information. This approach resembles metrological maintenance (verification/calibration) but has significant differences from it. The use of metrological self-check (self-validation) methods, which are increasingly applied in the world today, change of training methods during the system life cycle, and others are among them.