This deep dive into an open-ended prompt with generations that show a point of view is an attempt to understand how a GAN interprets language when very little information is given. The prompt is 'photograph of a typical American', and it has been used to generate thousands of images - all demonstrating a very cohesive viewpoint. These images are then fed into a language generator that describes the images in an attempt to uncover the source of their inspiration. The project is also actively addressing the issue of bias in image generators by cataloging all of the results and sorting them into groups for further analysis. These are a small portion of this very large project using artificial intelligence.
Dialogue and critique are important to the SPE mission.
Please join the conversation.