AI's Colonial Archives
Abstract: Generative AI technologies, such as text-to-image models, have received a lot of attention recently. As with all AI technologies, critics note these models enact various forms of simplifications, erasure and bias in their outputs. Yet, to understand the visual representations AI models produce (and to disrupt them), we must understand the politics and history of the canonical archives they build upon, attending to the power differences and gazes that have historically been amplified in these archives. In this paper, we evaluate AI-generated images by situating them within broader global histories of cultural preservation and representation of visual archives. We specifically focus on the representations AI models generate of the global south. By juxtaposing AI-generated images about communities and practices from the global south with historic examples from visual archives we can trace the lineage of the racial, ethnic, gender and class narratives these models reproduce. We use critical scholarship on colonial archives, museum curatorial practices, and history of photography, to show how the visual archives underpinning AI models are sites of miscategorization, produced through an elite imperialist understanding of the “other” to perpetuate an orientalist gaze. Importantly, this gaze persisted in the cultural archives produced not just “by the west” but from within the south also, complexifying the question of where we can find an ‘inclusive’ archive.
Author bios: Rida Qadri is a Research Scientist at Google. Her research interrogates the overlaps between culture and AI. She is interested in the organizational, epistemological and geographical cultural assumptions underpinning the design and deployment of AI systems. She also studies the tensions and frictions that emerge when mono-cultural AI design choices are universalized through a global deployment at scale. Prior to joining Google she completed her PhD at the Massachusetts Institute of Technology in Urban Studies.
Huma Gupta is Assistant Professor in the Aga Khan Program for Islamic Architecture at MIT. Gupta holds a PhD in the History and Theory of Architecture and a Master's in City Planning from MIT. Currently, she is writing her first book The Architecture of Dispossession, which is based on her research examining state-building through the architectural production of the dispossessed. Her broader research interests include the economic, cultural, and political relationships between discourses of architecture, development, and urban planning. Developing methodologies using sonic, visual, and other sensory archives to construct histories of subaltern spaces and subjects is of particular interest to her.
Fuchsia Hart is Iran Heritage Foundation Curator for the Iranian Collections at the Victoria and Albert Museum in London. She holds a BA in Persian, an MPhil in Islamic Art with Arabic, and is working towards the completion of a PhD at the University of Oxford, also in Islamic Art, with a focus on shrines in 19th-century Iran and Iraq.
Katrina Sluis is Associate Professor and Head of Photography & Media Arts in the School of Art & Design at Australian National University where she convenes the Computational Culture Lab. Katrina’s research is broadly concerned with the politics and aesthetics of art and photography in computational culture, its social circulation, automation and cultural value. As a curator and educator, for the past decade she has worked with museums and galleries to support digital strategy, digital programming and pedagogy. Her present work addresses the emerging paradigms of human - machine curation, as a contemporary response to the massive intensification of global image production and circulation.
Emily Denton (they/them) is a Senior Research Scientist at Google, studying the societal impacts of AI technology and the conditions of AI development. Prior to joining Google, Emily received their PhD in machine learning from the Courant Institute of Mathematical Sciences at New York University. Though trained formally as a computer scientist, Emily draws ideas and methods from multiple disciplines, in order to examine AI systems from a sociotechnical perspective. Emily’s recent research centers on a critical examination of the histories of datasets — and the norms, values, and work practices that structure their development and use — that make up the underlying infrastructure of AI research and development
Recorded Presentation | 27 April 2023
#History #Colonialism #ArchivesAndMuseums #GenerativeAI
