Machine Reflections is a series of AI-assisted self-portraits that explore gender bias.

By using large-scale generative models to create portraits of people with specific qualities and traits, this piece investigates how these models were distorted by preconceptions built into the large datasets they were trained on. Based on gender-neutral prompts of varying complexity, like "a successful lawyer," "portrait of a person," "portrait of beauty," "the best professor in the world," "the best teacher in the world," and dozens more, the project collected a large set of machine-generated images that reveal the often heavily gendered and stereotypical nature of these systems.

The final five images in this series, Strength, Leadership, Power, Femininity, and Beauty are self-portraits created by looking into the machine as a mirror. Blending an artist-curated selection of machine-generated images with the artist's self-photograph, these images hope not only to highlight biases in large-scale machine learning systems, but also to serve as a reminder that the underlying datasets leading to these results are human-made collections of (often unfiltered) data from the internet, which is a collection of social, human activity in itself.

Machine Reflections was presented in an Artist Spoitlight talk at the 2022 NeurIPS Workshop on ML for Creativity and Design.