"Stereo-typing: LLM chatbots’ appearance of typing can increase belief in a false stereotype"
by Williams-Ceci, Sterling; Zalmanson, Lior; Macy, Michael; Naaman, Mor (2024)
Abstract
LLMs are now central to our information-seeking practices. While much research has focused on the information these models generate, less work has investigated how LLMs' presentation of this information affects users' beliefs. We study how the "Typewriter Effect"'--a typing animation popular in LLM interfaces--influences users’ acceptance of a false stereotype, particularly among individuals in the targeted identity group. We conducted an experiment (N=990) in which participants interacted with an LLM-powered chatbot which endorsed the false stereotype that men outperform women in mathematics, with or without using the typing animation. We found that the typing animation increased women’s belief in and willingness to share this stereotype. This effect was mediated by perceived anthropomorphism. Overall, we highlight an overlooked risk of this common design feature, propose an alternative to mitigate it, and argue for greater consideration of design choices’ cognitive effects on users when evaluating AI-generated misinformation.
Keywords
Themes
AlgorithmsLinks to Reference
Citation
Share
How to contribute.
