Building Trust in ML Systems: Effective Stakeholder Communication

In this talk, Joseph Tenini explores the crucial aspect of building trust in Machine Learning (ML) systems, specifically focusing on communication with non-technical users and stakeholders. He outlines four key pillars for achieving this trust: describing performance relative to an interpretable baseline, quantifying uncertainty in the delivery process, sharing the "why" behind non-binary decisions, and designing for second-order process effects. The presentation emphasizes the importance of meeting stakeholders where they are, managing expectations about system imperfections, and enabling them to become champions of the ML products by understanding their underlying logic and broader impacts. The talk draws on real-world examples and practical advice to equip ML practitioners and managers with the tools to foster trust and maximize the value of their work.