Abstract:

What does it mean to adopt an evidence-based (EB) approach in the design and evaluation of intelligence-augmentation interventions that utilize explainable AI (XAI) systems? This keynote will explore an Evidence-Based eXplainable AI (EB-XAI) framework that underscores the importance of grounding AI interventions in empirical research. The framework introduces an evidence strength scale, aimed at encouraging studies that use real-world, rigorously vetted datasets and involve actual practitioners in the evaluation process. We will present a set of design-oriented concepts, along with model- and performance-oriented metrics—such as data reliability, model utility, model calibration, model robustness, and the impact of models on decision-making—that provide a foundation for comparing and assessing XAI solutions. Additionally, I will demonstrate an online tool that introduces the concept of "reliance" in human-AI collaboration, moving beyond the notion of mere "use." The appropriateness of human reliance on AI will be examined from both epistemological and ethical perspectives, guiding the selection of specific human-AI interaction protocols to be deployed in work and decision-making contexts where human and AI systems operate together. Choosing the right interaction protocol can ensure better collaborative performance and promote greater human sustainability compared to alternative approaches.

Biography:

Federico Cabitza is an associate professor at the University of Milano-Bicocca (Milan, Italy) where he teaches human-computer interaction, information systems and decision support. He is head of the Laboratory of Uncertainty Models, Decisions and Interactions in the department of Informatics at the above-mentioned university and is director of the local node of the Italian national laboratory “Computer Science and Society.”

Since 2016, he has been collaborating with several hospitals, including the IRCCS Ospedale Galeazzi Sant'Ambrogio in Milan, Italy, with which he has a formal affiliation and founded the Medical Artificial Intelligence Laboratory. He is associate editor of the International Journal of Medical Informatics (Elsevier ISSN: 1386-5056) and a member of several editorial boards.

His research interests are in the design and evaluation of artificial intelligence systems to support decision making, especially in health care and law, and the impact of these technologies on the organizations that adopt them. To date, he has published more than 160 research publications in international conference proceedings, edited books and high-impact scientific journals and is listed among the world’s most influential scientists, according to Stanford’s Top 2% Scientists list. He is the author with Luciano Floridi of the book “Artificial Intelligence, the use of new machines” published by Bompiani.