03 The Creative Act
Generative AI models clearly have an important impact on the creative process of many artists, with activities required involving both continuity and rupture with previous creative processes. These shifts are most visible in the reorganisation of creative workflows around the generative process, conceptual shifts around the nature of machine learning (ML) outputs, and evolutions in artists’ embodied experience of their practice (Ploin et al. 2022).
Typically, AI-enabled arts processes are never an AI-only exercise; there is often a dialogue between analogue and algorithmic techniques shaped by human decision-making. The issue of latent space is also a question of artistic agency, encompassing not only the selection and preparation of training corpora for models but also the curation of potentially limitless visual and textual outputs. Both of these activities constitute subjective acts that are complementary to the automation enabled by the computer. The selection of training material and curation of outputs are factors that can ensure that works are not derivative and have some kind of traction in the real world. Despite the increased affordances of AI technologies, the relationship between artists and their media remains essentially unchanged, as artists ultimately work to address human rather than technical questions. According to a report from the Creative Algorithmic Intelligence Research Project regarding the production of synthetic images, “[w]hile ML models could help produce surprising variations of existing images, practitioners felt that the artist remained irreplaceable in giving these images artistic context and intention – i.e. in making artworks”, and in a more general conclusion, “that human/ML complementarity in the arts is a rich and ongoing process through which artists refract technological capabilities to produce artworks” (2022, 6). The ways of seeing and embodied knowledge that an artist brings to their practice are essential when curating outputs from GenAI systems. It is the artist’s repertoire of cultivated sensibilities shaped by aesthetic judgement, conceptual orientation, and historical awareness that guide meaningful selection and transformation. Rather than merely choosing from pre-generated options, the artist enacts a critical and creative process, recognising formal or thematic potential, drawing unexpected associations, and situating fragments within a broader narrative or conceptual structure.
This curatorial process is inherently performative, resembling the role of a film editor, choreographer, or composer who shapes disparate elements into a cohesive and intentional whole. Without such informed discernment, the outputs risk remaining generic, directionless, or detached from context. It is through this iterative engagement, grounded in the artist’s accumulated knowledge and creative intuition, that generative material is reframed and activated, allowing for works that carry depth, coherence, and cultural resonance. This reliance on repertoire connects closely to Marcel Duchamp’s concept of the “personal art coefficient” from his lecture “The Creative Act”, which describes the gap between an artist’s intention and what is actually realised in the artwork. Duchamp suggests that within this space, the unintentionally expressed becomes significant, revealing truths that escape conscious control. In the context of GenAI, the artist’s role becomes one of discerning within the machine’s output those unexpected, emergent moments that become the “unintended” expressions Duchamp saw as central to artistic creation. GenAI may produce endless variations, but it is the artist’s repertoire that enables the recognition of accidents, glitches, or surprising harmonies as holding aesthetic or conceptual weight. In doing so, the artist perpetuates the creative act not by making from scratch but by navigating the ambiguous space between control and chance, between authorship and reception that Duchamp saw as fundamental to art. By way of example, in the next section, 04 Summary of Project Method, it can be seen how a clip extracted from a British public information film of a man being deliberately tripped up has elicited the text, “it was nothing but a new syntax”. This text was only one of many generated outputs from the assemblage responding to that footage, but the selection of this particular text output was not arbitrary. Considering that the language model was trained on a corpus of Michel Foucault, the selected text seemed to crystallise a conceptual blend between the visual source
material, the logic of comedy, and Foucault’s philosophy of knowledge systems. Comedy often works by disrupting narrative flow, revealing the constructed nature of what we take for granted. Similarly, Foucault’s genealogical method examines how social institutions produce and legitimise systems of order through language, discipline, and categorisation. In this sense, an unexpected fall, perceived as a comedic moment, becomes a disruption in the continuity of meaning, exposing the grammar of social behaviour, while momentarily undoing its coherence. The generated phrase, “it was nothing but a new syntax”, operates as a poetic compression of this idea – a break in structure that reveals the possibility of reconfiguration. The curatorial act then, was not a matter of choosing a line that merely “fit” but one that enacted a synthesis between the disparate elements of gesture, text, and theory to generate new interpretive potential. Through this approach, the artist’s role becomes not only that of prompt engineer or editor but also a translator across systems – reading the output not as static content but as a hinge point between visual provocation and discursive reflection. The generative model offers raw material, but the work of shaping meaning rests with the artist’s capacity to sense alignment and tension between elements, and to recognise the possibility of critical insight within a moment of absurdity or error.