Agonistic Image Generation: Unsettling the Hegemony of Intention

A figure from my paper showing the different interfaces we tested. Our agonistic interface is in the bottom right!

I wrote this paper with my co-author, Andre Ye, advised by Drs. Ranjay Krishna and Amy Zhang from Winter 2024 to Spring 2025. In this paper, we argue that image generation interfaces should be more agonistic (a term drawn from Chantal Mouffe's account of agonistic democracy), meaning that they should push back against user intentions by engaging users with political controversies over visual representation. We built an interface that uses AI to research alternative intepretations of user prompts from Wikipedia, and found through a lab study that the interface is more effective than others at facilitating critical reflection. This is my first computer science paper and it was accepted to the 2025 ACM Conference on Fairness, Accountability, and Transparency in Athens!

The inspiration from this paper came in Winter 2024 when I was participating in a faculty cohort meeting with Dr. Zhang. In the cohort meeting, we discussed a recent controversy over Google's Gemini AI model generating pictures of racially diverse Founding Fathers and Nazis. As a philosophy student, the ethial dimensions of this story immediately interested me, so Dr. Zhang suggested I talk to Andre, who was also a philosophy and computer science student. Coincidentally, Andre and I also happened to be taking the same Jewish philosophy class (PHIL 418). Although I knew that Andre was "the other CS and philosophy guy," I hadn't talked to him much before and didn't know much about him. I approached him on the last day of the class, and to my surprise learned that he had also been looking for a collaborator to start a research project!

Over the next year, I spent many hours building the interface and conducting user interviews, all while juggling responsibilities like an Amazon internship and classwork. At times, I hated working on the code for the project, while at other times, I loved the interdisciplinary nature of the work. All the hard work paid off, however, once we learned that we were accepted to the conference!

Writing this paper has had a transformative impact on my post-college career plans. Before starting the project, I had no intention of pursuing further education in computer science and planned to go straight into a career in tech policy, or perhaps law school. The image I had of computer science researchers involved either low-level hardware engineering or abstract theoretical mathematicians, neither of which particularly appealed to me. This project opened my eyes to the world of human-computer interaction (HCI) research, which I learned could be an ideal place for me to bring together my interests in technology and philosophy. While I am still interested in tech policy, I am now considering pursuing my tech policy interests through a computer science graduate program, which would allow me to continue exploring questions like the ones I'm asking in this paper.

If you're interested in reading the paper, I've attached a preprint here:

Agonistic_Image_Generation_Final.pdf

Download PDF • 10.16MB