Hybrid talk with Margaret E. Roberts (University of California, San Diego)

Increasingly popular and capable large language models (LLMs) have become a key source of information for millions of people around the world and have the potential to shape public opinion. However, the information provided by these models is not always neutral, but can be politically biased or perceived as biased. We analyse how people’s preferences for LLM output and their perceptions of political bias in LLM output are affected by partisanship. We run a conjoint experiment in the U.S. where respondents are shown different model responses that vary in terms of (1) the prompts they respond to—aligned with the respondent’s party or aligned with the opposing party—and (2) whether the LLM hedges, refuses to answer, or provides a two-sided response to the prompts. We test whether there is a “preference gap”—differences between parties in internalised preferences about the degree to which LLMs should moderate its output—and “party promotion”—a preference for the LLM refusing to answer with politically misaligned content. We also test whether the amount of LLM refusal affects trust in LLMs and generative AI companies.

The event is the last of four lectures taking place in the frame of the Summer Lecture Series of the SCRIPTS Forum.

Information on the speaker:

Margaret Roberts is a Professor in the Department of Political Science at the University of California, San Diego. She co-directs the China Data Lab at the 21st Century China Center and is an affiliate at the UC Institute on Global Conflict and Cooperation. She is the recipient of the 2022 Max Planck-Humboldt Award and holds a Chancellor's Associates Endowed Chair at UCSD. Her research interests lie in the intersection of new technologies and digital politics, with a specific focus on the politics of artificial intelligence, online censorship and propaganda, and science and innovation.

Roberts received a PhD from Harvard in Government (2014), MS from Stanford in Statistics (2009) and BA from Stanford in International Relations and Economics (2009). Her first book, Censored: Distraction and Diversion Inside China's Great Firewall, published by Princeton University Press in 2018, was listed as one of the Foreign Affairs Best Books of 2018, was honored with the Goldsmith Book Award, the Best Book Award in the Human Rights Section and the Best Book Award in the Information Technology and Politics Section of the American Political Science Association. Her second book, Text as Data: A New Framework for Machine Learning in the Social Sciences (published with Justin Grimmer and Brandon Stewart) won the American Sociological Association’s Methodology Section’s Outstanding Publication Award in 2025.