Developing Standards for AI/ML Systems in Civil Aviation: Challenges and Barriers

The inability to establish appropriate assurance methods for AI/ML components in safety critical systems leaves us unable to effectively manage the risks and benefits of such systems. It drives cost of development for systems with AI/ML components uneconomically high, it delays the adoption of systems with AI/ML components at scale, and it can result in catastrophic consequences in terms of the safety of systems with AI/ML components. In this presentation we will explore what constitutes sufficient scientific-based evidence to substantiate a safety claim related to an AI/ML component performing a safety-critical function.