The introduction of generative AI technologies has the potential to transform creative production by allowing for the rapid creation of media objects with little human input. But if aspects of production can be automated, will “machine creativity” lead to a decline in content diversity? In this workshop, we overviewed the harms of the creative use of AI, such as bias and homogenization, and shared practical techniques to re-imagine AI as "a serendipity machine” that supports divergent creativity. See the slide deck for materials.
As a part of the workshop, participants used Stable Diffusion to generate images as a way to probe the inherent biases and homogeneities of the system. This interactive webapp shows the 55 images generated over the two workshop instantiations. The images are plotted in a 2D "latent space" using UMAP projections of CLIP embeddings. Zoom and drag to explore the space and hover over images to read image prompts.
We need to harness public intelligence to steer AI systems towards ethical uses. An important part of that is crowdsourcing critique and understanding of AI generated images, which can help us understand the underlying biases and homogeneity that exists in these generators. For example, we discovered that images of barracudas often contain humans holding up a prized fish, reflecrting a latent human cultural practice that has implications of AI-imagery (see slide 6 of the slide deck or The Barracuda Effect on Wikipedia).If you want to help with the citizen science of exploring this latent space, get in touch or check back for a interaction way to add images to the map.
Links and resources: