Social Lunch ~ Gandalf Nicolas ~ Rutgers University

Gandalf Nicolas, Assistant Professor of Psychology, Rutgers University

 ~ Linguistic Social Cognition in Humans and Artificial Intelligence Language Models.

Abstract ~ In this talk I will present my research on linguistic social cognition in humans and Artificial Intelligence (AI) language models (e.g., ChatGPT). In a first set of studies, I present the Spontaneous Stereotype Content Model (SSCM), a high-dimensional taxonomy of linguistic stereotypes. Human subjects open-endedly provided characteristics they associate with salient social categories (e.g., gender, race, occupations), and responses were coded using AI text analyses. The SSCM reveals a diverse set of dimensions and properties associated with stereotypes in language. A second set of studies expands this approach to intersectional targets (i.e., those with multiple salient social categories). Open-ended stereotypes reveal that intersectional targets are associated with more diverse, unique, and negative stereotypes, as compared to their constituent single-categories, underscoring the nuances of complex categorical perceptions. Finally, I present research showing how AI language models, trained on vast amounts of human data, learn and reproduce human stereotypes. I show how AI language models can teach us about cultural patterns, and how psychological theory can inform research on fairness in AI to improve auditing and debiasing frameworks, reducing harmful consequences of biased emerging technologies.

Bio

Gandalf Nicolas received his Ph.D. in Psychology at Princeton University and is currently an Assistant Professor of Psychology at Rutgers University - New Brunswick. He studies how people think about the social world and how Artificial Intelligence (AI) models learn and reproduce these social representations. His research is at the intersection of social cognition and interdisciplinary quantitative methods (e.g., computer vision and natural language processing). Specific topics include linguistic social cognition, intersectional perceptions, social biases in AI, and human-computer interaction.