Undergraduate Dissertation: Social Biases in Generative Artificial Intelligence
My undergraduate dissertation on gender bias in generative AI: what “unbiased” should mean, and how a philosophical framework can guide practical alignment.
in Generative AI
The question
What does it mean to ask a generative AI system to be unbiased? My undergraduate dissertation examines the difference between representing the world and reproducing its social prejudices, with a particular focus on gender.
The philosophical framework
The dissertation develops an “Ideal World” model: a world as close to our own as possible, while excluding social prejudices that are irrelevant to the work being generated. It explores the tensions between ethical alignment, historical accuracy and user intent.
From theory to practice
The second half evaluates data filtration and augmentation, prompt manipulation and fine-tuning against that framework. It considers human and AI feedback as ways to connect a philosophical account of bias with practical model development.