Robust and Explainable Measurement of AI Political Bias
Navn på bevillingshaver
Dustin Wright
Titel
Postdoctoral Fellow
Institution
University of Cambridge
Beløb
DKK 1,695,290
År
2025
Bevillingstype
Internationalisation Fellowships
Hvad?
This project asks: how politically biased is artifical intelligence (AI), how does this bias manifest itself in AI systems, and how does this impact real people?
Hvorfor?
From asking questions to ChatGPT, to seeing AI generated content online, people interact with AI systems every day. If these systems show persuasive political bias, for example by strongly favoring particular political issues, they can worsen other societal issues, such as polarisation. Identifying these biases, their causes, and their impacts, will help us to develop and use AI responsibly.
Hvordan?
This project will develop a new method to uncover AI political bias that is more robust and explainable than previous methods. It will look at what arguments AI systems present in favor or against different political issues in order to measure the underlying biases hidden behind these arguments. It will then examine the data the systems are trained on in order to explain the presence of any bias.