(in progress project) Computer American Dreams is an art book of generative images of U.S. history. I wanted to explore the idea that our dreams are shaped by our pasts, and that our futures cannot escape our histories. I used a contemporary technology (artificial intelligence) trained on archival image datasets to generate new images. The project explores how the things we imagine are informed by the patterns and traumas of our pasts and how technology is a form of memory.
I used an AI trained on archival imagery to generate computer “dreams” based on historical data. I scraped and cleaned the custom dataset from various government and newspaper archives documenting the 20th century. The images on this page were generated from photographs of WWII, which often showed soldiers in close-up or marching in a large group or military camps. Training on archival images was difficult because of the poor quality of old photos, their limited supply, and their wide variation in subject matter, position etc. My initial models were prone to mode collapse, but I was able to get the model to converge by using a KNN classifier to pre-categorize the images and other optimization techniques.
The project grew out of my desire to explore the context-based nature of available information, the conflicts embedded within facts, with the idea that facts themselves can be carriers of injustice. The project is set to be completed by April 2019.