UIUC Quant Brownbag
“Pictures in our heads”: Revealing hidden biases in face representation via deceptively simple tasks
Stefan Uddenberg - University of Illinois Urbana-Champaign
When we look at a face, we rapidly and automatically form impressions of traits like trustworthiness, competence, and dominance — judgments that can have real-world consequences. My work investigates the visual roots of these social psychological processes, revealing the hidden biases embedded in our face representations through deceptively simple experimental paradigms. First, I will discuss how iterated learning — a method akin to the children's game of “Telephone” — demonstrates how stereotypes of facial trustworthiness can emerge and persist even in the absence of any real-world correlation. Second, I will show how generative models of human faces enable us to empirically map the trait impressions people form based on facial appearance, allowing for precise manipulation of high-level attributes to explore novel research questions. Finally, leveraging a deep learning-powered reverse correlation technique made possible by my earlier work, I will explore how people's mental representations of a “good leader” are shaped by their political attitudes — revealing systematic differences with potential implications for elections and leadership selection. Together, these findings highlight how our perceptual system is tuned to extract social meaning from faces, and serve to illustrate how integrating cognitive science, social psychology, and machine learning can unlock new insights into the biases that shape our social world.