서울대학교

[정치외교학부 10-10 사업단]

SEOUL POLITICAL SCIENCE AND INTERNATIONAL RELATIONS SERIES

Can AI Reduce Prejudice at Scale?

CHARLES D. CRABTREE (Assistant Professor, Department of Government, Dartmouth College)

일시: 2025년 4월 22일(화) 15:00-16:30
장소: 223동 우석경제관 504호
* 행사는 모두 영어로 진행됩니다.
* 사전 신청없이 참석 가능합니다.

Authors: Charles Crabtree, John B. Holbein, Mitchell Bosley, Semra Sevi
Abstract: Recent research has demonstrated that personalized interpersonal contact can reduce prejudice and promote support for marginalized groups. However, scaling these interventions is challenging because they require significant resources to train human facilitators and conduct in-person conversations. This study examines whether artificial intelligence (AI) can offer scalable, personalized, and persuasive interventions to increase support for transgender rights. Building on research that suggests moral alignment enhances persuasion and that AI chatbots can help reduce conspiracy beliefs, we create an AI-powered messaging intervention aimed at U.S. respondents during the 2024 presidential election. Using OpenAI's GPT-4o, we conducted personalized conversations that referenced the moral values of participants. Our findings show significant short-term increases in support for transgender rights, with Cohen's d estimates ranging from 0.11 to 0.21 across various measures. Additional analyses confirm that these results are robust to a range of potential validity threats. However, follow-up results are more variable, suggesting that the long-term impact of the interventions remains unclear. We conclude that AI-driven personalized persuasion can lead to meaningful short-term effects, but further research is needed to understand how lasting these changes are and what mechanisms drive them.

문의처: 정치외교학부 10-10 사업조교(lveronica93@snu.ac.kr)