We present experiences from evaluating the learning performance of a user-adaptive personal assistant agent. We discuss the challenge of designing adequate evaluation and the tension of collecting adequate data without a fully functional, deployed system. Reflections on negative and positive experiences point to the challenges of evaluating user-adaptive agent systems. Lessons learned concern early consideration of evaluation and deployment, characteristics of AI technology and domains that make controlled evaluations appropriate or not, holistic experimental design, implications of 'in the wild' evaluation, and the effect of AI-enabled functionality and its impact upon existing tools and work practices.

Neil Yorke-Smith is an Assistant Professor of Business Information and Decision Systems at the Suliman S. Olayan School of Business, American University of Beirut, and a Research Scientist at SRI International, USA. His research interests include planning and scheduling, intelligent agents, preference modelling, hybrid OR/MS methods, machine learning and data mining, and their real-world applications.