Development of an Effective Safety Assessment Capable of Measuring the Safety of AI-Based Systems.

AI-based systems are highly advanced and data-driven employed to analyze, predict, and monitor their surroundings, and finally make informed decisions within a specific context. The development of a specific class of Artificial Intelligence (AI)-based systems known as safety-critical systems (SCS) demands a heavy emphasis on core system-wide attributes such as availability, safety, trustworthiness, robustness, and reliability. We study existing verification tools and methods that have been utilized to improve the reliability and trustworthiness of these types of systems, together with their limitations. There are different challenges experienced in each of the stages of building SCS making it difficult to create a safe system. If the systems are built without addressing the challenges, they end up having issues themselves. To certify these systems, one needs to understand all current issues that arise due to the challenges, and if possible, solutions to them are sought. In this paper, we contribute to the certifying board by providing a possible method S(D2TV) to use in the certifying process. We conclude with new directions to address the limitations of SCS verification and stress the importance of developing a safe AI-based system.