Are AI Models Politically Neutral? Investigating (Potential) AI Bias Against Conservatives

Shalevska, Elena and Walker, Alexander (2025) Are AI Models Politically Neutral? Investigating (Potential) AI Bias Against Conservatives. International Journal of Research Publication and Reviews, 6 (3). pp. 4627-4637. ISSN 2582-7421

Text IJRPR40238-Are AI Models Politically Neutral.pdf - Published Version Download (319kB)

Abstract

This empirical, mixed-approach research investigates whether different popular Generative AI models such as ChatGPT-3.5, ChatGPT-4, Gemini, and Microsoft Copilot show political bias, particularly against conservative viewpoints. The study uses the popular political affiliation test – the Political Compass – to assess the political leanings of the AI's responses. By analyzing the Political Compass Test scores and the individual output provided by the AI models, the study finds that the models’ responses consistently align with more liberal, left-hand standpoints, indicating a potential bias against conservative ideologies. While recognizing the role users play in shaping AI outputs, the study emphasizes the importance of these models providing a neutral baseline. Neutral AI systems are critical to ensuring balanced political discourse, avoiding the amplification of existing inequalities, and supporting informed democratic engagement. These findings contribute to the ongoing discussion about the impartiality and fairness of AI systems in political discourse

Item Type:	Article
Subjects:	Scientific Fields (Frascati) > Humanities > Languages and literature Scientific Fields (Frascati) > Humanities > Philosophy, ethics and religion Scientific Fields (Frascati) > Social Sciences > Political science
Divisions:	Faculty of Education
Depositing User:	MA Elena Shalevska
Date Deposited:	27 Mar 2025 09:03
Last Modified:	27 Mar 2025 09:03
URI:	https://eprints.uklo.edu.mk/id/eprint/10820

Actions (login required)

View Item