Synopsis

In the summer of 2020, I began exploring her genealogical journey through a personal and an AI lens. I interviewed family members and began following the migrational patterns of my ancestors across Arabia and Mesopotamia spanning five generations. 

AI Failures

While researching and collecting digital archives relevant to my family's stories, I discovered that the only visual records lacked indigenous self-expression, and emanated from a British Empire that purposefully crafted the East (i.e., the "Orient" or "Other") to be exploited and controlled. As I tested these images with computer vision AI models, this process revealed “failures” manifested as mislabeling, generalizations, and stereotypes. They failed to recognize the majority of veiled women as women. They haphazardly tagged several bedouin images with modern-day warfare labels such as "soldier, "army," and "military uniform," confidently asserting high confidence values in their evaluations. These “failures” uncovered not only the prejudice systemically-embedded within commercial AI tools, but the broader problematic results that arise from using a historical training dataset that lacks an understanding of "Middle Eastern" imagery.

Salaf symbolizes my frustrations with the Western colonial gaze and the lack of native localized self-expression. I deliberately erased the “oriental” stereotypical figures from my datasets, using an AI segmentation technique called U-2 Net, and trained an AI on this “absent” dataset. These deliberate reversals of technology's functions symbolize the erasure of my ancestors' collective memory while confronting AI's reduction of history and identity.

Featured

Slanted Magazine #37—AI