P1 guest talk by Responsible AI researcher Alexandra Olteanu

Title: (Un)rigorous AI

Abstract: Beliefs and assumptions about what AI systems can, will, or should do abound. Although often unsupported and sometimes questionable, these beliefs and assumptions have had a formative impact on how AI systems have been built, evaluated and described. This in turn impacts how and in what settings they are deployed and used, and thus when they fail. However, foreseeing and addressing the consequences of putting insufficiently rigorous AI artifacts (e.g., models, systems, applications, outputs, data) into practice are often seen as the purview of responsible AI—which is generally seen as separate from rigor, as it is understood as more concerned with ethics, stakeholders, societal impacts and harms, and real-world deployment scenarios. I argue that what responsible AI asks researchers and practitioners is to uphold principles of scientific integrity in their work. And, although responsible AI is often seen as out of scope for an AI researcher or practitioner not engaged in that space, any scientist should see rigor as well within scope. In this talk, my goal is to provide a useful framework for broadening our collective understanding of what rigorous AI research and practice should entail (including how we communicate about AI systems), and to help scaffold much-needed dialogue about and scrutiny of current AI system development and evaluation practices.

Bio: Alexandra Olteanu is a Responsible AI researcher. Until recently she was a principal researcher at Microsoft Research Montreal, part of the Fairness, Accountability, Transparency, and Ethics (FATE) Montreal team; which she was a founding member of and later rebuilt. She is also an associate industry member at Mila. Her work routinely unpacks and challenges practices and assumptions made when designing, evaluating, deploying or communicating about AI systems. Her most recent work focuses on 1) rigor in AI (with a focus on conceptual clarity, measurement, and evaluation), and 2) human agency (with a focus on anthropomorphic AI systems’ design, behavior, and impact). Before joining Microsoft Research, Alexandra was a Social Good Fellow at IBM's T.J. Watson Research Center. Alexandra has co-organized tutorials/workshops and has served as PC/SPC/AC and in many other service roles for most major web, social media, and responsible AI conferences. She has also served as the inaugural co-chair of the Fairness, Accountability, Transparency and Ethics on the Web track @ The 2023 Web Conference, and as the 2024 Program Co-chair for the ACM Conference on Fairness, Accountability, and Transparency. Alexandra holds a PhD from École Polytechnique Fédérale de Lausanne (EPFL), Switzerland.