"Person” takes the process of ethnographic representation and turns it on its head: the text of fieldnotes from an ethnographic encounter is used to generate a synthetic set of images representing a single informant’s struggles with air pollution. We see 64 representations of ‘Franck’ – a Martinique-born Parisian, father of two, avid runner, and transport manager – filtered through multiple layers of interpretation and representation, from the ethnographer to GPT4 to Midjourney to the artists. 

While they fail as portraits – Franck is physically never the same – these representations take us into Franck’s everyday realities and to impossible places – episodes in his past, distant places outside the ethnographer’s reach, his dreams. In composite, they are a powerful affective portrait of how Franck encounters a polluted world: the experiences that shaped him, his anxieties and nervous habits, and his efforts to cope. 

Yet these representations also obscure his experiences. To ‘paint a picture’ of his life, they rely on pastiche, collapsing familiar corners into globally famous landscapes; they misrepresent his physical self and race; they over-dramatize subtle or ambiguous experiences with established styles and iconography. The ethnographer’s eye for the complexity of everyday life is lost. 

While some are absurd misunderstandings or hallucinations, many are gaps in precision met with leaps in interpretation, or gaps in data met with unknowing substitutions. As we collaborate more with AI systems, is this affective essence enough? Will we need greater ethnographic precision to make AI tools useful? Will models need to map the world better to reflect an ethnographer’s nuance?