Künstliche Intelligenz: Bild und Bias
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Philipps-Universität Marburg
Abstract
Image-generative AI produces distorted, often stereotypical images of reality. Why is this the case and how can it be counteracted? In the presentation, the functionalities of Generative Adversarial Networks and diffusion models are presented with advantages and problems in the use, bias in existing models is shown and strategies for avoiding bias are discussed. The focus is on the generation of gender-equitable image material, whereby intersectional aspects are also taken into account.
The presentation accompanies episode 13 of the podcast AI in Teaching on bias in image-generative AI:
https://www.youtube.com/playlist?list=PLLmr_XhQwwKNAuJLA8bpNUHv0QiGvvs1v