one day a person created an image, which they wanted to share with other people.
they posted it online, and people who saw it knew who made it. it had attribution.
millions of other people did the same.
for years people posted images all over the web.
billions of human-made images were uploaded.
one day, these images were scraped into datasets intended for research.
scraping removed people’s identities from the images. attribution was lost.
only the images remained.
one day all these images were used to train a powerful AI.
anyone could use this AI to create images just by typing a few words.
every image it made was influenced by the human-made images it was trained on.
stable attribution is finding the human-made images and the people who made them, for every image the AI makes.