Annotation Fever! (2025)
“Just as the entire mode of existence of human collectives changes over long historical periods, so too does their perception. The way in which human perception is organized—the medium in which it occurs—is conditioned not only by nature but by history.”
Walter Benjamin’s insight frames how photography enters generative AI: emerging media instigates new patterns of perception, which are stabilized by their incorporation into practices, instruments, and institutions that, in turn, reorganize the sensorium for new modes of perception that are at once individual and collective, organic and technical.
In today’s generative AI, the photographic is everywhere and nowhere at once. Photography’s histories, conventions, and archives form the very foundation of training large datasets, which, in turn, condition the emergence of statistical logics used to read and generate new sets of images. At the core of these processes lies annotation. Annotation is central to documentary practice. In On This Day, this becomes evident in the Beijing Silvermine archive of vernacular Chinese photography, where the date stamps dotting negatives function as a kind of pre-AI annotation. These marks are not simply placeholders but sites of events: ruptures in representation that cannot be fully translated yet insist on acknowledgment. They highlight both what is present and what resists circulation. Annotation is also the basis of today’s multimodal models, which rely on tens or even hundreds of thousands of dispersed human annotators, meaning that every machine-readable image bears a collective intelligence. For these works, to annotate is to learn to see as the machine sees, aligning human intuition with statistical deduction. These annotations and the images they produce increasingly permeate collective vision. Today, virtually no image remains untouched by some aspect of algorithmic parsing, segmentation, and classification. Our everyday visual culture is conditioned by how models sort and generate images; even a phone snapshot is shaped by countless micro-mediations trained on ideals of photographic appearance.
The exhibition and workshops expose these invisible negotiations of labeling, segmentation, and inscription while asking what needs to be withheld, erased, or left unmarked. The resulting images redirect attention to questions of authorship and ownership: to whom does the annotated image belong, and should it be circulated, donated for further training, or withheld from computational afterlives? In the gallery space, images marked with a red sticker dot indicate consent for further computational use. To annotate a model is not only to break an image into machine-legible units; it is to fine-tune the future terms of vision, marking what tomorrow’s collective perception will become.”
Klara Källström & Thobias Fäldt, Thomas Sauvin, Bernard Dionysius Geoghegan, 2025