Salaf [Arabic: سلف , ancestor] (2020) is part of my ongoing exploration into constructing my genealogical journey using two different voices: my own and an AI ‘narrator’.
As part of my residency at ThoughtWorks Arts exploring synthetic media, I began following the migrational patterns of my Saudi and Iraqi ancestors leading up to my own migration process to the US. I interviewed family members and studied the socio-cultural events transpiring in their distant memories and anecdotes. Through this process, I traced my lineage to the 1800s and began experimenting in re-imagining the past using Generative AI models. As I juxtaposed my story with the viewpoints of Western AI systems, I encountered a narrative widely different from my personal one. The models failed in recognizing the nomadic bedouin faces in my stories and even regurgitated familiar stereotypes and cliches about the Arab world. In one particular experiment, it tagged most of my images with arcane modern-day warfare labels such as “soldier, “army”, and “military uniform”. AI generalized how it sees me and generations of family before me. I discovered that my collection of stories were being erased and stereotyped as one big pile of “Middle East”.
Salaf symbolizes those AI failures and the frustrations I felt in the Western colonial gaze, and the lack of native localized self-expression. Using U-2 Net, an image segmentation model that partitions a digital image into multiple segments, I erased the ‘oriental’ stereotypical figures in my historical archives, creating an “absent” dataset. I then trained StyleGAN2, a generative AI model on this new dataset, outputting images signifying the eradication of her ancestor’s collective memory
Back To The Gallery