Presentation
Bias Emergence in LLM-Generated Narratives: Structured Probes of Social Class and Race Stereotyping
SessionPoster Session 1
DescriptionAs AI becomes a more visible part of everyday life, understanding how it may unintentionally reflect bias is critical. This study explores how large language models like ChatGPT handle scenarios involving race and class. In first experiment, we test whether a suspect’s background (wealthy vs. poor) affects how the AI discusses guilt. In second experiment, we analyze how it describes characters from diverse racial and national groups. Using a mix of human evaluations and computational tools, we assess whether stereotypes appear in the model’s responses—and how much randomness in generation influences that. Our results highlight the challenges of building fair and trustworthy AI, especially when used in sensitive or high-stakes contexts. We offer practical insights for developers and designers seeking to mitigate AI bias through both technical improvements and human-centered design.
Event Type
Poster
TimeTuesday, October 14th5:30pm - 6:30pm CDT
LocationRiverside East