Exhibiting AI images: Unexpected outputs, linguistic inequalities, and curatorial possibilities

Abstract

Structured around latent data spaces that store masses of semantic associations, AI image generators have an intrinsic capacity to draw connections their users might not expect. Their outputs thus have the potential to engender characteristically anthropological epistemic shifts: subverting pernicious stereotypes, opening up horizons of possibility, and rendering the familiar strange. I demonstrate this with reference to World Hypnosis Convention , a collage of AI images that I developed for a public exhibition following long‐term fieldwork on the social imagination of hypnosis. While English‐language prompts generated highly standardized portraits of male stage performers, other languages elicited vivid, unexpected portraits that challenged rigid definitions of “hypnosis,” drew attention to the constitutive role of suggestion in social life, and invited critical perspectives on social institutions. World Hypnosis Convention thus demonstrates how AI images can support practices of anthropological exhibiting, while illuminating several limitations of AI image generation itself. In particular, it highlights the linguistic inequalities with which AI users can be faced: non‐English speakers may struggle to achieve the precision they may desire, whereas English speakers can miss out on lateral perspectives. Both issues warrant greater attention in critical anthropologies of AI and in future AI design.