A position on the emerging debate
There is a debate that has been circling itself for three years, and which in recent months has begun to find a partial resolution. I intend to explain why this is a partial resolution, and to propose what I consider to be a complete one.
The debate is as follows: who is the author of a work created with artificial intelligence? For three years, responses have oscillated between two poles, and both have become exhausted.
At one pole lies legal anthropocentrism. Copyright offices—led by that of the United States—require a 'controlling' human behind the work to grant it protection. Without a conscious human subject in command, there is no author and no protectable work. It is an administratively convenient and philosophically impoverished position: it reduces the question of authorship to a search for the finger that pressed the button.
At the other pole lies co-authorship through the sum of parts. An AI work has no single author, it is argued, but several: the model programmer, the dataset, the individual who writes the prompt. Distributed authorship. It is less naive than the former, yet it remains trapped in the same question: it merely replaces 'one author' with 'several authors'. It continues to seek subjects to whom the work may be assigned.
Between both poles, in recent months, a third, more interesting path has begun to emerge: returning to Alfred Gell. Gell’s anthropology of art—Art and Agency, 1998—possesses a virtue that the debate required: it deactivates the requirement of consciousness. For Gell, an object possesses agency not because it 'thinks', but because it acts within a network of social relations: it fascinates, compels, captures. The work is a 'trap' that retains the intention of its maker and acts upon the viewer. Applied to AI, this allows us to speak of the agency of the work without attributing a mind to it. It is a step forward, and I understand why the field is turning in that direction.
However, it is a partial resolution, and it is necessary to explain why.
Gell resolves one problem while inheriting another. He resolves the issue of consciousness: we no longer need to ask whether the machine thinks. Yet he inherits a vulnerability that is well-documented within anthropology itself. The classic critique of Gell—formulated by Robert Layton, Howard Morphy, and others—is that his notion of object agency is 'an analogy taken too far': objects do not act, nor do they change of their own accord; rather, there is a subject who attributes agency to matter. The Gellian trap only functions insofar as it effectively simulates human intention. That is to say: Gell displaces anthropocentrism, he does not transcend it. He maintains it by deferred means. The AI work would be a trap because it successfully imitates what a human would do.
Furthermore, there is something else, which is decisive. Gell's ontology is instrumental at its root: for him, art is a system of action designed to influence others—to capture, to enchant, to exert social control. The trap requires functional performance: it exists to do something to someone. This aligns poorly with what truly interests us when we speak of art, and it aligns particularly poorly with AI, where the primary function of the system is already defined in advance: to generate an output that responds to the prompt. If art is merely effective capture, then an AI image that fulfils its brief well would be the perfect case of Gellian art. And we know that is not the case. Fulfilling the function does not make it art.
This is where I have been working in a different direction for years, and where I believe the complete resolution lies.
What defines the artistic event is not who originates it—human or machine—nor its efficacy in influencing others. It is something I have termed structural surplus: a relational difference that persists and reorganises the field of a system's possible configurations, without being absorbed by the primary function of that system.
Translated to the context of AI: the primary function of a generative system is to produce an optimal output that responds to the prompt. That is the purpose for which the system exists. The artistic event—if there is one—does not lie in the fulfilment of that function, however well it may be performed. It lies in the surplus: in what remains, in the difference that the system produces which its own function fails to reabsorb. Art is not the successful output. It is the remainder that overflows the task.
This shifts the entire question. It ceases to matter who the author is—the question of the first two poles. It ceases to matter whether the work 'acts' as a trap—Gell's question. The question becomes: does this encounter produce a surplus that reorganises something and cannot be reduced to the function that originated it? And that question can be asked equally of an oil painting, a generative work, or even an interaction between two machines, without the need to seek a subject behind it or to pretend that the object possesses volition.
The advantage is not rhetorical. Gell explains one part well: how AI functions as an infrastructure of capture that redistributes prior human agencies—those of the engineers, those of the millions of artists whose work is contained within the data. That is true and useful. But Gell cannot conceive of what we do need to consider: the possibility of an artistic event that does not refer back to any human agent, not even in a deferred manner. The structural surplus can, because it does not locate art in the agent that originates the action, but in the persistence of a difference that no agent controls entirely.
I write this because the debate is currently taking shape, and because I believe it deserves not to remain at its halfway point. Returning to Gell is a sound step. But it is a step that keeps art tethered to function and agency to the mimicry of the human. The question of art in the age of AI is not whose work it is, nor whom it captures. It is what, in that encounter, overflows the function and reorganises the field.
That is what I have been calling structural surplus, and it is the framework from which I shall continue to work on this matter. Not as another variant of the relational turn, but as its consequence: if art was always relational, AI inaugurates nothing new. It only renders undeniable that the artistic event was never in the author, but in what remains.
This notebook remains open, as always, to anyone who wishes to discuss it.
Sources
Gell, A. (1998). Art and Agency: An Anthropological Theory. Oxford: Clarendon Press.
Layton, R. (2003). Art and Agency: A Reassessment. Journal of the Royal Anthropological Institute, 9(3), 447–464. https://doi.org/10.1111/1467-9655.00158
Morphy, H. (2009). Art as a Mode of Action: Some Problems with Gell’s Art and Agency. Journal of Material Culture, 14(1), 5–27. https://doi.org/10.1177/1359183508100009
Van Eck, C. (2010). Living Statues: Alfred Gell’s Art and Agency, Living Presence Response and the Sublime. Art History, 33(4), 642–659. https://doi.org/10.1111/j.1467-8365.2010.00767.x
Esteban Ruiz, J. A. (2026). Art as Structural Surplus: Toward a Relational Ontology Beyond Human Authorship (V2.3). PhilArchive and Zenodo.
Esteban Ruiz, J. A. Processes Without Meaning. Public notebook at juanesteban.art, 2026.