Bridging the risk awareness gaps between the safety science and AI domains: A perspective with engineering applications

In recent years, significant advancements have been observed in the artificial intelligence (AI) domain with a growing adoption in various safety-critical engineering applications. However, numerous AI-related incidents indicate that its intelligence cannot guarantee its safety. This has raised the need for safe and risk-aware AI, enabling deeper understanding of system and operational risk. Efforts have been put in the AI domain to tackle the challenge of "risk awareness", but with a considerable difference from the safety science domain. Therefore, the aim of this article is to identify and assess differences in how risk awareness is understood and used in the safety science and AI domains, focusing on engineering applications. A systematic literature review has been conducted to a) analyze the use of the risk awareness concept in the safety science domain and synthesize a generic description, b) analyze the risk awareness concept in the AI domain, and c) assess the differences between these concepts to identify the gaps. The result highlights the need for proper integration of safety science knowledge for safe AI solutions to be developed and achieved. Furthermore, this study provides a way forward towards future research directions for risk-aware, safe, and trustworthy AI agents.

Keywords: Artificial intelligence, autonomy, safety, risk, risk awareness, risk aware AI.