Ghosts & Urgent Futures
Concluding a Research Project & Talking About Resistance
Building on my previous two posts as a Flickr Foundation research fellow, I’ve concluded my writing series (and my fellowship) with part three of The Ghost Stays in the Picture. Whereas parts one and two focused on the path of images into datasets and then into infrastructure, part three examines the ways that infrastructure becomes “haunted” by what enters into it.
It’s an admittedly messier analysis, and I shift gears to look at the ways certain words in a dataset become associated with particular histories. Specifically, I look at the word “stereoview” as a prompt for generative AI models and attempt to read them, winding my way through the history of stereoview images and the surrounding cultural context of American imperialism in the Philippines. It is not only the technological apparatus that we conjure up when we ask Midjourney for examples of these types of images — it is also the cultural context in which the original images were produced.
You can read part one here and part two here, or jump straight to part three with the button below.
Caroline Sinders and I were interviewed for Jesse Damiani’s Urgent Futures podcast, which you can watch on YouTube (above) or find as a podcast pretty much everywhere, including Apple Podcasts and Spotify or wherever else through Reality Studies. It’s a great chat about “Glitching AI, Algorithmic Resistance, Labor Activism, Art as Research, & Feminist Technology.” What else could you ask for in a podcast?
Good Reads This Week
Ryan McGrady and Ethan Zuckerman write about YouTube, AI and training data this week in The Conversation, and it’s worth a read. They present their research “taking a closer look at some of our more surprising findings to better understand how … obscure videos might become part of powerful AI systems. We’ve found that many YouTube videos are meant for personal use or for small groups of people, and a significant proportion were created by children who appear to be under 13.”
YouTube is a rough place for researchers to study, as are many platforms, and any insight into the training data for AI systems is hard-won. It’s worth reading what they’ve shared.
