=================
University of Gothenburg, Sweden
Explainability as an Evaluation Problem
Bio. Alan Said is an Associate Professor of Computer Science at the University of Gothenburg, Sweden. His research focuses on human-centered AI, recommender systems, and user modeling, with an emphasis on evaluation, transparency, fairness, and sustainability in AI-driven systems. His work explores how personalization systems can be designed and assessed in ways that are both effective and socially responsible. He earned his Ph.D. from TU Berlin on recommender system evaluation and held a Marie Curie Fellowship at CWI, postdoctoral position at TU Delft, alongside experience in industry. He has authored over 100 publications and received awards including the Springer Best Paper Award at UMAP. He currently serves as Chair of the ACM RecSys Steering Committee.
Abstract. Explainability has become a central goal in personalized and adaptive systems, driving the development of new explanation algorithms, interfaces, and, more recently, LLM-based approaches. Despite this progress, there remains limited understanding of what makes an explanation effective and how its impact should be assessed. Explainability is often treated as a feature to be added rather than an outcome to be evaluated. This talk argues that explainability is fundamentally an evaluation problem. Drawing on lessons from recommender systems research, it examines how reliance on weak proxies, limited reproducibility, and insufficient user-centered evaluation can obscure whether explanations genuinely increase understanding, improve trust, or support better decision-making. By revisiting how we evaluate explanations and their effects, the talk highlights both longstanding challenges and new opportunities for explainable user modeling.