Keynote: Operationalizing Trust for Human-AI Collaboration

Abstract

What does it take for people to genuinely trust AI? Principles like accountability, transparency, and fairness are important, but they do not tell us how AI should behave in interactions with humans. In this talk, I approach this question through sociotechnical systems, where humans and AI agents collaborate by combining their distinct capabilities. For such systems to be effective, agents must exhibit and understand skills such as empathy, self-reflection, and social awareness—skills typically associated with humans. Thus, we must develop ways to represent, measure, and evaluate these skills within computational models. Privacy provides a particularly rich domain for examining these issues. I will present our work on privacy assistants that reason about norms, contexts, and values to help people make informed choices in complex data-sharing environments. This work demonstrates how abstract notions of trustworthy AI can be operationalized into concrete mechanisms for human-AI collaboration.

Bio

Pınar Yolum is Professor of Trustworthy AI at Utrecht University, where she leads the Responsible AI research group within the Department of Information and Computing Sciences. Her research focuses on multiagent systems and human–AI collaboration, with an emphasis on trust, commitments, and privacy. Pınar serves on the Editorial Boards of Journal of Autonomous Agents and Multiagent Systems, ACM Transactions on Internet Technology, and IEEE Internet Computing. She is the program leader of the AI Master’s program at Utrecht University, co-founder of the Open-Source Global Justice Investigations Lab, and a member of the Executive Board of the Hybrid Intelligence Consortium. She regularly contributes to activities to promote women participation in computer science. She is an enthusiastic reader and a restless skier.