Min-Yen Kan
National University of Singapore, SG
Fact checking is dead. Long live fact checking!
The race between misinformation generation and verification is becoming increasingly unequal. AI can produce and adapt misleading content faster than reactive fact checking can investigate and correct it. This keynote advances a polemical argument: if our response remains checking claims after they circulate, misinformation wins. As AI mediation becomes ubiquitous across how we learn, communicate, and decide, the information retrieval community must help society get ahead of this cycle. This challenge demands both proactive systems and a more informed, critical citizenry. Bringing people along requires making critical inquiry part of everyday encounters with AI-mediated information. People need the capacity to question how AI frames an issue, selects evidence, omits perspectives, and presents uncertainty. Where did this information come from? What supports it? What is missing? Whose interests and assumptions shape what I see? Systems must support this inquiry by modelling both information and its value to human understanding and decision-making.
Drawing on insights from the NUS iGyro project, this keynote outlines a vision for the next stage of AI-mediated information retrieval and its implications for the community’s research agenda: helping citizens question, evaluate, and shape the mediation on which they increasingly depend.
Jing Jiang
Australian National University, AU
Evaluation has been a central topic in information retrieval, natural language processing, and more recently generative AI. Although task performance remains a key concern in evaluating large language models, a user-centric approach requires consideration of many other factors and dimensions. In this talk, I will discuss some of these considerations, including bias, safety, and culture, with a particular focus on the evaluation of vision-language models.
