An Assurance Case Practice of AI-Enabled Systems on Maritime Inspection

Assuring of AI-enabled systems is challenging and beyond current assurance practices, especially in bridging the gap between assurance process and tools. In this paper, an AI-powered corrosion detection system for maritime inspection is presented as an assurance use case. It serves as a decision support tool for surveyors to assess the coating conditions in ballast tanks. Before deployment, it is crucial to establish confidence of this system as the stakeholders seek to understand the potential risk of adopting it compared to existing or alternative solutions. Different from other works focused on assurance process or framework, this paper conducts both assurance process and testing methods to create a detailed assurance case. A systematic top-down approach is used to derive the assurance requirements from the system level to the machine learning component level, while testing is conducted using a bottom-up approach to collect the required evidences. Furthermore, a risk-based approach is integrated into the corresponding AI assurance lifecycle, providing valuable insights on analyzing the risk that an AI component may bring into AI-enabled systems.