DiSCourse Seminar with Indira Sen
19 June 2026, 12:00 (CEST), hybrid
Digital Science Center, Innrain 15, 1st floor, Open Space Area or Big Blue Button
DiSCourse* - The Digital Science Seminar Series on:
Measuring Political Values in Large Language Models
Since people seek help from Large Language Models (LLMs) in many social contexts, e.g., political advice, auditing LLMs’ efficacy includes studying their political values and becomes an interdisciplinary challenge. In this talk, I first focus on measuring LLMs' political slant by repurposing survey instruments developed for people and using NLP methods to measure partisanship in LLM responses. However, LLMs’ partisan slant tells us little about their tendencies regarding cross-cutting issues like democracy. In the second part, I discuss the degree to which LLMs uphold democratic values through a systematic computational audit. We filter large-scale anonymized datasets of human-LLM conversations to find prompts where people ask LLMs about democracy-related tasks or advice, e.g., "what’s the use in voting when all politicians are corrupt?" We then evaluate LLM responses using a validated classifier that detects reinforcement versus pushback on antidemocratic tendencies. This audit sheds light on how LLMs can improve or erode democratic literacy.
*featuring a distinguished guest: Indira Sen, University of Mannheim
Indira Sen is a Junior Faculty member at the University of Mannheim’s Business School, where she is part of the Chair of Data Science for the Social and Economic Sciences. Her research sits at the intersection of Natural Language Processing (NLP) and Computational Social Science. She investigates the societal impact of language technologies, including Large Language Models, and explores how humans interact with AI systems.