“we design and deploy a facilitated ex- perience where participants collaboratively speculate on utopias they want to see, and then produce AI-generated imagery from those speculations.”
participants paired up and responded to this prompt: “How will we re-imagine the following categories in utopia?”
Next, we took the 32 visions, and ran them through VQ- GAN+CLIP, a common model of text-to-image synthesis.
This produced visual representations of the participants’ vi- sions, and served as the output of the speculation experience.
Evaluation through semi-structured user interviews
First, to measure relatedness between prompt and image, we tested if participants could recognize the image gener- ated by their prompt
Regardless of their answer, we then re- vealed the correct image.
We then asked questions related to their interpretation of the image, starting by asking them to describe what they saw. Based on directions from this ini- tial description, we then asked follow-up questions about if and how the image differed from their expectations.
Finally, we asked if the participant had ideas for a follow-up prompt that better reflected their vision.