AI systems malfunction — and when they do, a human is typically held responsible for identifying and correcting this malfunction, yet they are rarely trained to do so. In this talk, I will advocate for AI malfunction drills: training exercises that would expose participants to simulations of realistic AI system failures in a controlled setting. These exercises would provide a safe environment for the users and operators of an AI system to practice detecting and resolving AI malfunctions.
Rosco Hunter is a PhD student at the University of Warwick with a background in Maths and Computer Science. His PhD research focuses on how automation can streamline the design and governance of AI systems (e.g., using automation for hyper-parameter optimisation and scalable risk management).
