Assuring the Case: A Safety Engineering Approach to AI-Enabled Systems

Given the increasing ubiquity of Machine Learning (ML), there is an urgent need for assurance frameworks, particularly in relation to ML-enabled safety critical systems. We begin by describing the process of assurance case construction for conventional safety-critical systems that has been developed by the safety engineering community. Goal Structured Notation (GSN), a graphical approach to assurance case construction, is then briefly described. In the sequel, we describe an existing ML assurance framework that employs GSN, Assurance of Machine Learning for use in Autonomous Systems (AMLAS), which is a six-stage assurance process. AMLAS is then applied to our use-case, namely Autonomous Vehicle (AV), with several of the AMLAS stages described in detail. In particular, the creation and testing activities of the ML assurance stage and the instantiation of the assurance argument are developed. Finally, we make the case for increased interaction between the ML and safety engineering community.