AI and safety management

Artificial intelligence based on deep learning, along with big data analysis, has in recent years been the subject of rapid scientific and technological advances. These technologies are increasingly being integrated into various work environments with the aim of enhancing performance and productivity. This dimension of the digital transformation of businesses and regulatory authorities presents both significant opportunities and potential risks for industrial safety management practices. While there are numerous expected benefits, such as the ability to process large volumes of reliability data or unstructured natural language incident reports, the structural opacity of large neural networks, their nondeterministic nature, and their capacity to learn from new data mean that traditional safety assurance techniques used for conventional software are not applicable. Additionally, the expansion of the scope of automatable tasks and the gradual move towards work collectives that are composed of human operators who collaborate with various intelligent machines and agents introduce new variables that must be considered alongside and integrated with the organizational and human factors of safety.

What are the main challenges posed by these new technologies in terms of skills management, worker well-being, privacy protection, and the pursuit of performance that aligns with societal expectations? What changes are required in how we conceptualize the safety of high-stakes activities, how we demonstrate and verify the absence of unacceptable risks, and anticipate potential deviations?

This document provides a concise overview of the most recent available information, contextualized by decades of research on automation in high-hazard systems. It focuses specifically on the projected impacts for high-hazard industries and infrastructures over the next ten years.