Artificial Intelligence (AI) is advancing at a pace never seen before.The technology is now being rapidly adopted at scale in all potential cross sections of society, while the safety implications are not being considered.This work presents a systematic literature review on the elicitation of safety requirements on a combination of AI and traditional software from the automotive domain.We focus on the era before Generative Pretrained Transformer (GPT).Automotive is one of the industries that has adopted AI at scale with technologies such as adaptive cruise control and lane assist, which are leading to autonomous driving.There are many parallels to the current rapid development in AI in the automotive world.These include the move towards automated driving, high competition, frequent release cycles and the price-sensitive nature.These characteristics set the automotive industry apart from other safety-critical industries.Ensuring safety in the automotive industry starts with safety requirements.A plethora of safety requirements elicitation processes and techniques for the automotive domain have been proposed.To the best of our knowledge, no study characterizes them.This article characterizes the state-of-the-art in eliciting safety requirements via a systematic literature review.We selected 102 primary studies from 2097 related articles.We identified and compared nine distinct processes for safety requirement elicitation.We constructed taxonomies of 38 distinct techniques used to conduct the different steps in each of the processes.This article can act as a guide and 'cheat sheet' for beginners and practitioners to choose processes and techniques for their projects.For researchers, this study provides an overview of the field, research gaps and future research opportunities.
Kochanthara et al. (Thu,) studied this question.