Assurance Guidance for Machine Learning in a Safety-Critical System

Abstract

We are developing guidance for space domain assurance personnel on how to assure Artificial intelligence (AI) and Machine Learning (ML) systems. Key to such guidance will be an assurance process for these personnel, who may be unfamiliar with such systems, to follow. We are investigating one such process, the “Assurance of Machine Learning in Autonomous Systems (AMLAS)” from the University of York, UK. To gauge its suitability, we are (retrospectively) applying it to a safety critical AIIML system in the space domain. We report here on our experience so far in applying this process.