JohnWillis
New member
- Joined
- Mar 9, 2026
- Messages
- 9
I've been obsessively reading about potential thesis topics and found gold at the University of Zurich's political science department. They're offering several MA thesis topics related to AI and politics for 2026 . Sharing in case others are as interested as I am. 
Topic 1: Ideological Bias in Large Language Models
This explores whether AI systems carry implicit worldviews. The central challenge is defining what "political bias" even means — is it about outputs? Training data? The cleaning and filtering procedures? How alignment and fine-tuning affect ideological orientation .
Topic 2: How AI Systems Are Used in Practice
Beyond technical design, how are these systems used in research, media, governance, and lobbying? What does it mean for democratic accountability when information intermediaries become algorithmic ?
Theoretical vs. Empirical approaches:
Topic 1: Ideological Bias in Large Language Models
This explores whether AI systems carry implicit worldviews. The central challenge is defining what "political bias" even means — is it about outputs? Training data? The cleaning and filtering procedures? How alignment and fine-tuning affect ideological orientation .
Topic 2: How AI Systems Are Used in Practice
Beyond technical design, how are these systems used in research, media, governance, and lobbying? What does it mean for democratic accountability when information intermediaries become algorithmic ?
Theoretical vs. Empirical approaches:
- Theoretical: Clarify what it means to speak of "ideological bias" in AI, compare to bias in media or survey design.
- Empirical: Test different models on survey-like or political prompts, analyze variation across providers, explore how alignment affects ideological orientation .
- Do different LLMs have consistent political leanings?
- How do alignment procedures affect political outputs?
- Can we detect ideological bias in supposedly neutral systems?
- How are AI systems shaping political discourse?
- Argyle et al. (2023) — Using language models to simulate human samples
- Bisbee et al. (2023) — Synthetic replacements for survey data
- Boelaert & Ollion (forthcoming) — Machine Bias: Generative LLMs Have a Worldview of Their Own
- OpenAI (2024) — Defining and Evaluating Political Bias in LLMs