Dr. Anke Stoll (University of Amsterdam, Netherlands) Human Judgments, Machine Decisions: Annotation Bias in Machine Learning Classification (Cancelled)

In this talk, I discuss my ongoing research on bias in machine learning systems, focusing on how systematic differences in human judgments may shape the decision-making of models trained on these judgments. In particular, I explore how insights from the social sciences can inform research designs for detecting and modeling annotation bias in machine learning systems. I present work from my current project at the University of Amsterdam with colleagues from communication science and social psychology, where we examine whether the political orientation of annotators influences how AI models classify facial expressions (i.e., emotions). I also discuss my work on annotator bias in the detection of uncivil and deliberative comments in online discourse, showing that decisions about what constitutes a valuable user contribution can be influenced by the formal education of annotators.

As machine learning systems increasingly operate within digital media environments, they may shape how people are represented, categorized, and moderated in mediated communication contexts. I look forward to discussing how research on AI bias can connect with perspectives and research questions in communication science.