DEBORA HIRSCH: CATALOGUES AND WRITINGS

My dataset did not begin as a dataset. It began as an accumulation of images: my own paintings and photographs, rare books and old maps, tapestries, vases, azulejo panels, pelourinhos, colonial churches, Baroque architecture, and digital interventions gathered across different places and periods. Before these images became material for artificial intelligence, they belonged to a personal archive shaped by years of travel, research, artistic practice, and sustained looking.

An archive is never simply a place where images wait. The act of collecting already establishes relationships among things that were previously separate. A photograph of a colonial church encounters a fragment of a map; a painted plant appears beside an ornamental motif; the geometry of an azulejo panel enters into conversation with the architecture of a pelourinho. Each image carries its own material history, but its position within the archive gives it another life, determined by proximity, recurrence, and association.

When these images enter a dataset, their condition changes again. The dataset is not merely a digital container for the archive. It translates images into information that can be read by a computational system. Color, contour, texture, composition, and spatial relationships become part of a field of numerical correspondences. Objects separated by geography and historical time may become connected through formal resemblances that neither their makers nor their collectors originally intended.

At the center of my process is a pretrained artificial intelligence model that I fine-tune using a dataset assembled from this personal archive. The distinction is important. Large generative models are commonly trained on immense quantities of images gathered from the internet, often without meaningful attention to origin, authorship, historical context, or cultural specificity. My dataset is deliberately limited and situated. Its contents have been selected because they belong to the history of my practice and to the questions that have accompanied it.

The model therefore works within a visual field that I have constructed. It encounters my earlier paintings alongside photographs of colonial architecture, decorative objects, botanical materials, maps, and historical artifacts. It identifies patterns and generates variations from relationships embedded in that field. The resulting images are not retrieved from the archive, nor are they simple copies of individual sources. They emerge from statistical correspondences among many images, filtered through the particular limits of the dataset.

This process introduces a peculiar form of historical compression. Objects made centuries apart, under radically different social and political conditions, may coexist within a single generated image. A botanical figure may acquire the structure of a Baroque ornament. The surface of a vase may become architectural. A fragment of azulejo may migrate into a plant, a landscape, or an invented body. Historical distance is not erased, but reorganized into new and sometimes unstable visual relationships.

The archive from which these images emerge is inseparable from the colonial histories embedded within it. Churches, maps, pelourinhos, decorative arts, and botanical illustrations were not passive witnesses to colonial expansion. They participated in systems through which territories were measured, populations governed, plants classified, resources inventoried, and power made visible. Beauty and violence often occupied the same object. The refinement of an ornamental surface could coexist with the coercive structures that financed or surrounded its production.

To work with these materials is therefore to enter a history that cannot be approached as pure visual inheritance. Their forms remain compelling, but they are also marked by the conditions under which they were produced and circulated. In my work, artificial intelligence becomes a means of returning these images to a state of instability. It loosens their established arrangements and permits fragments of architecture, ornament, botany, and landscape to enter configurations that no longer confirm the order from which they came.

The model does not interpret this history by itself. It does not understand colonialism, memory, or loss. It produces formal possibilities from the material on which it has been trained. Meaning enters through the construction of the archive, the selection of the dataset, and my subsequent decisions about which outcomes to preserve, reject, combine, transform, paint, or animate. Artificial intelligence expands the field of possible images, but authorship remains located in the choices that establish the field and give the final work its form.

This alters the conventional chronology of artistic production. The archive is no longer consulted only before the work begins, as a source of reference or research. It remains active throughout the process. It informs the dataset; the dataset conditions the model; the model generates variations; and those variations return to the artist as material for further selection and intervention. The work moves repeatedly between collection and invention, machine generation and human judgment, historical evidence and speculative form.

The process also changes my relationship with my own previous work. Paintings and images that once appeared complete can reenter the archive as sources for new compositions. Their forms may be divided, displaced, or recombined with materials that preceded them by centuries. The archive ceases to describe a linear history of finished works. It becomes recursive. Earlier images contribute to later ones, while later images change how the earlier archive can be seen.

This recursive movement complicates the idea of originality. The resulting compositions do not originate from a single gesture or source. They emerge through layers of transmission: an object is made, photographed, collected, digitized, incorporated into a dataset, interpreted computationally, selected, transformed, and returned to material or moving form. At each stage, the image acquires something and loses something. Authorship resides not in the fantasy of creation from nothing, but in the deliberate construction and navigation of these transformations.

The use of blockchain introduces another understanding of the archive. A traditional archive preserves through physical custody. Its objects remain vulnerable to decay, dispersal, institutional decisions, and historical catastrophe. A blockchain record uses distributed verification to make retrospective alteration difficult, giving the digital work a persistent and traceable history. It does not release the work from time, but it records the work’s passage through time differently.

This is especially significant in projects concerned with endangered species, damaged ecosystems, and colonial memory. The blockchain can preserve an image and document its continuity, but it cannot preserve the living organism or restore the historical world to which an artifact once belonged. Its persistence reveals the distinction between keeping a record and protecting what the record represents. The digital trace may remain precisely because the material or biological referent has become precarious

My archive therefore operates in several forms at once. It is a collection of objects and images, a history of my artistic attention, a dataset for computational generation, and a structure through which cultural memory can be reorganized. It contains materials inherited from systems of classification and power, but it also allows those materials to be detached from their established hierarchies and placed into new relations.

The archive is not a neutral storehouse, and the dataset is not an innocent technical instrument. Both are structures of selection. They determine what enters, what remains absent, what can be connected, and what kinds of images may become possible. By constructing the dataset from my own archive, I make those choices visible as part of the work.

Artificial intelligence does not replace the archive. It activates its latent correspondences. It reveals how one image may persist inside another, how historical forms migrate across time, and how the acts of preserving and transforming are never entirely separate. What emerges is not a recovery of the past, but a new space in which its fragments can meet, resist one another, and acquire an unfinished future.