Ting LUO (University of Birmingham)

Abstract:

Large language models are increasingly used in decisions that affect people's lives, including what political content millions of people see, whose loan applications get approved, and which job candidates advance, decisions that reflect underlying political values, not neutral calculation. Globally, the landscape of large language model development is dominated by two countries: the United States and China. The political values embedded in these systems are shaped through opaque processes of model development and deployment, with limited democratic input or public accountability. When these models are adopted across different contexts, what political values do they carry with them? This talk discusses early results from a research project that aims to answer this question. We designed a battery of questions about governance trade-offs and queried 130 open weight large language models these questions. We test 1) whether these models give complete, consistent answers at all or whether they are essentially just guessing; 2) whether their answers differ between models built by Chinese and American developers.

Short Bio:

Ting Luo is an Associate Professor in Government and Artificial Intelligence at the Centre for Artificial Intelligence in Government, University of Birmingham. Her research sits at the intersection of technology and governance, with a particular emphasis on digital governance and how digital development shape politics and governance. Her work has been funded by the British Academy, UKRI and the National Institute for Health Research. She holds a PhD in government and an MSc in public policy and administration from LSE.