A discreet task has occupied museums, archives, and universities for years: converting art into data. Not the images, which is already accomplished, but the knowledge surrounding the works. Who painted what, when, with which materials, who attributed it and upon what basis, what it represents, what it signifies. All of this is being encoded into formal structures, ontologies, and knowledge graphs, so that machines may relate, search, and cross-reference it across collections worldwide. An international model exists for this purpose, including an ISO standard, alongside projects that model the creative process of a work or the levels of interpretation within an image.

This is a valuable endeavour. It would be facile to dismiss it as technocratic reductionism, yet that is not the case. The transition of a painting's attribution from a mere loose phrase to a formal assertion accompanied by a date, author, and source—one capable of coexisting with competing attributions without erasing them—represents a genuine advancement for research. The ability to cross-reference a museum record from Tokyo with an archive in Seville, ensuring they speak the same language, is of no small consequence. I do not come to mock the graph.

I come to inquire what it preserves and what it omits, for not everything surrounding a work of art can be contained within a node. It is often asserted that art cannot be codified because it possesses an ineffable quality, a mystery that eludes reason. This response is convenient and, in my view, false: it transforms the problem into mysticism and thereby closes it. What the graph fails to preserve is not a mystery inherent in the material of the painting. It is something of a different nature, and for that reason, it is not a datum.

A knowledge graph effectively records everything that can be asserted about a work of art: its author, date, provenance, iconography, and even the various interpretations it has received and who held them. At its best, it is an exhaustive map of what can be said about the piece. And what can be said is substantial, and worthy of record. However, what makes a work art is not a property that can be noted in a field, no matter how refined that field may be. It is what occurs when someone encounters the work and something in their way of seeing is reorganised. That does not reside in the work as pigment or date do: it resides in the relationship between the work and the viewer, and a relationship is not inscribed; it is either produced or it is not. The graph may store the fact that a painting moved generations; it cannot store the movement itself, because that movement is not an attribute of the painting, but an event that the painting makes possible.

I am not concerned that the graph is incomplete, as it will always be, but rather with what happens when what fits into the graph begins to be valued more than what does not. Every data structure makes choices: it decides what constitutes an entity, what constitutes a relationship, what deserves its own field, and what remains unrecorded. These decisions are not neutral, and those working in the sector know this better than anyone. A researcher in the field articulated this precisely: metadata does not speak for itself, and its structure, almost always invisible, determines what meaning can be extracted from it. The Getty vocabularies themselves warn that they must be able to accommodate the debated and ambiguous nature of information about art, rather than converting every doubtful attribution into a firm fact. This is a lucidity inherent to the field itself: the discipline understands that to encode is to decide.

The risk is not that the machine will misinterpret art. It is more subtle: that which does not lend itself to codification will gradually become invisible. What has no field is not searched for, and what is not searched for ceases to be found. If knowledge about art begins to circulate primarily through these structures, then what they cannot accommodate—open interpretation, fertile disagreement, what a work does that has yet to be articulated—runs the risk of being treated as noise, as that which has not yet been formalised, rather than what it truly is: the living part of the matter.

Let us not confuse the map with the territory and, above all, let us not cease to care for the territory because the map is becoming increasingly refined. A knowledge graph is a magnificent tool for knowing what is known about a work. But what a work is, in the sense that concerns me, is not what is known about it: it is what it continues to do each time someone stands before it. That does not fit into any node, not because it is sacred, but because it is not a thing. It is a relationship, and relationships are not stored. They are lived, or they do not occur.

On open conversation

This text examines a real and ongoing endeavour: the codification of knowledge about art into formal ontologies and graphs, using international models and projects that extend to formalising the creative process or iconographic interpretation. I am not opposed to this work, which is useful for research; I question what is stabilised when interpretation becomes a data structure, and what is left out. The distinction I propose, between what can be asserted about a work and what a work does before the viewer, stems from my research into recognition and surplus. If anyone wishes to intervene from the fields of digital humanities, heritage documentation, ontology engineering, or art history, this notebook remains open.

Sources

CIDOC CRM (ISO 21127:2023), a conceptual model for the integration of cultural heritage information, developed by the ICOM CIDOC.

Swiss Art Research Infrastructure (SARI) and its Creative Process Representation Ontology (CPRo), for the modelling of creative processes.

ICON, an ontology for iconographic and iconological interpretation based on Panofsky's levels (Sartini et al.; Baroncini, Daquino and Tomasi, 2021).

Canning, Erin, et al. 'The Power to Structure: Making Meaning from Metadata Through Ontologies'. KULA, 2022.

Getty Vocabularies (Art & Architecture Thesaurus), regarding the debated and ambiguous nature of art-historical information.

FOIS 2026 (16th International Conference on Formal Ontology in Information Systems, Vitória, Brazil), as the context for the field of formal ontology.