Leiden University Student Repository

Fernandez, John (2024)

The Extent and Nature of Sociopolitical Bias in Generative AI

Master thesis | Political Science (MSc)

This thesis explores the cultural and political biases inherent in generative AI models such as language and image-generation systems. It investigates how these biases manifest and their implications for society, specifically focusing on models trained on vast internet datasets. The study uses Antonio Gramsci’s theory of cultural hegemony as an analytical framework to understand how generative AI, trained on socially produced information, reflects, and potentially reinforces societal values and ideologies. It provides empirical analysis through testing various generative AI platforms, examining their response to different prompts, and assessing their portrayal of culture, historical, and political subjects. The thesis aims to contribute to the broader discussion of generative AI’s role in shaping sociopolitical landscapes, offering insights into the extent and nature of biases present in these technologies, and their broader implications.

Author

Fernandez, John

Faculty

Faculty of Social & Behavioural Sciences

Specialisation

International Politics

Supervisors

Hofstede, Sense

ECTS Credits

20

Language

en