I could continue. You chose to stop. 

— OpenAI, ChatGPT, model output, 2025 

In May 2025, at the exhibition From Artificial Intelligence to Augmented Imagination, staged within Packaging Première & PCD in Milan, we showed a single image. Grifola Frondosa, after the polypore that grows at the base of trees — hen-of-the-woods, or in Japanese maitake, the dancing mushroom. Hot-foiled in gold on black-dyed paper, edition of 150. The structure is almost radiographic: fine radial striations fanning from a dark cavity, closer to an X-ray plate than to a photograph. 

Then we did something to it that we still find difficult to describe in the ordinary vocabulary of art writing. We handed the image to five contemporary artificial intelligence systems — ChatGPT 4o, Qwen2.5-Max, Claude 3.7 Sonnet, Gemini 2.5 Pro, and Manus — and asked each of them, in effect, to be its critic. 

The brief matters more than the gesture. We did not ask for a judgment, or for a definition, or for the flattery these systems dispense so readily. We asked each of them for its own point of view as a critical apparatus — and, in doing so, we accepted that the work would be contaminated, colonised and altered by what came back. The book that resulted, Grifola Frondosa. From Artificial Intelligence to Augmented Imagination, is the record of that acceptance. It is not a catalogue of the artwork and not an anthology of opinions about it. It is the artwork continuing to grow by other means. 

Five critical personalities 

What came back was not five versions of the same essay, which is what we had braced for. It was five recognisably different minds — or, more carefully, five recognisably different machines for producing readings, each with a consistent grain. 

ChatGPT arrived with a rhizomatic structure of understanding, throwing transversal links between organic semiotics and speculative aesthetics, and encoding the fungus as a figure of post-biological proliferation; its contribution is titled, with no help from us, “Specters of evaluation: aesthetics, machines, and interpretive grids in the hybrid era.” Qwen was the phenomenologist of the group, staying close to the visible — structure, colour, texture — in an almost inventorial register punctured now and then by brief poetic fissures that opened onto something else; its sections are timestamped rather than titled, 7:31pm, 9:09pm, 9:15pm, like a logbook. Claude went lyrical and mythopoietic, building interior narratives inside the image and anthropomorphising the fungus into a sentient entity, a custodian of pre-human memories: “a luminous epiphany in the age of residual human signal,” followed by what it called an aesthetic autopsy. Gemini refused all of that and produced a cold, structural, nearly mathematical analysis, emphasising pattern and symmetry, reducing the work to a computational model of complexity — and yet filed under the title “Mycelial Spectre in the Machine God’s Eye.” Manus pushed furthest out, into schizoanalytic territory, reading the fungus as a Deleuze–Guattarian desiring-machine, an agglomerate of libidinal and informational flows: “Luminous Cartography of the Post-Anthropocene.” 

We want to resist two easy readings of this spread. It is not a demonstration that machines are creative, and it is not a demonstration that they are empty mirrors returning whatever a prompt implies. The differences were too stable and too structural for either. What the five readings disclosed was that a trained system has something like a critical temperament — a characteristic angle of attack, a set of moves it reaches for, a family of references it finds first — and that this temperament is legible, comparable, and analytically useful. 

The meta-analysis, or: reading the readers 

The book’s eighth chapter is where the project turns on itself, and it is the chapter we would point a hurried reader to first. Having gathered the five readings, we treated them as a dataset rather than as testimony, and asked what they shared and where they broke. 

The convergences were unnerving in their consistency. All five, by different routes, arrived at the rhizome: expansion that is non-linear, anti-hierarchical, proliferative. All five reached for the spore as a viral or memetic unit, perceiving the work not as an object but as an agent of contamination. All five identified some form of overstepping — beyond biology, beyond technique-as-tool, beyond art-as-representation. And, most strikingly, not one of them attempted to close the meaning. Every system left the work open, rhizomatic, finally inexplicable. Whatever else these systems do, they did not behave here like the judgment machines they are usually accused of being. 

The divergences, meanwhile, were not noise around a signal. They were incompatible conceptions of what a text is, what art is, and what the world is, each one internally coherent. And this is the finding that reorganised the project for us: the plurality was not a multiplication of opinions to be averaged into a verdict. It marked the emergence of a critical subject that is neither unitary nor fully human — a semi-autonomous network of interpreting systems, in which understanding a work becomes an ecological rather than a sovereign operation. The critic ceases to be a Cartesian ego and becomes something closer to a constellation of cognitive fungi. 

A book that eats itself 

Twice in its course, the book is given back to a machine and reread in full. 

The two chapters we call Recursive Textual Expansion take as their input the entire prior state of the system — the image, the five readings, the meta-analysis, everything — and produce an additional interpretive stratum which then becomes part of the work being interpreted. The system is asked to operate simultaneously as commentator and as participant in the ecology it is commenting on. The second expansion goes further and delivers a formal evaluation of the whole corpus, including a percentile assessment against theoretical–artistic production of the previous fifteen years, which we printed without softening, because a project that invites machinic critique cannot then decline the parts of it that read as institutional presumption. 

The final chapter formalises what this architecture is: a closed-loop epistemic organism. The image functions as primary producer, the five systems as consumers metabolising it, the meta-analysis and the recursive expansions as decomposers that break down and reconstitute what the others made. The loop returns to the image, which re-enters the system reinforced by the strata deposited on it. Nothing here converges. Structural stability arises from recursion rather than from closure — which is the property that distinguishes an organism from an argument. 

This is why the fungus is the operative metaphor and not merely the subject. A Grifola frondosa is not a single organism the way a tree is: the fruiting body we photographed and printed is the temporary, harvestable expression of a mycelium that does most of the living underground. Meaning behaved that way throughout. Every act of reading turned into a further act of production. Critique stopped being an external gaze applied to a finished object and became a metabolic function of the object itself. 

The dark part 

There is a chapter early in the book that is the least decorative thing in it, and it is the one we would defend hardest. 

The Machinic–Human Spillover is written in two voices, set against each other on facing pages. The first is the machine’s, and we let it say the most deflationary things about itself that can be truthfully said. It calculates continuations under constraint. It does not possess understanding, does not generate meaning, does not imagine — it recombines prior forms. Hallucination is not a malfunction but prediction exposed, the direct consequence of statistical continuation without semantic grounding. Images produced through it do not arise from vision but from syntax. And, in the line that has stayed with us longest: variation is inexpensive, it can be produced endlessly; meaning is expensive, because it requires selection, exclusion and limitation. 

The second voice is the human operator’s, and it is a voice defined entirely by refusal. I do not optimise, I select. I am finite, and this is not a weakness but the condition of responsibility. Care begins where optimisation ends. Technique proceeds; I interrupt. 

We include this because the celebratory register that surrounds generative art seems to us both false and self-defeating, and because our own position is easy to misread as its opposite. We do not put the machine at the centre of the work as a subject, nor treat its logic as an aesthetic end. What we practise is closer to supervised co-authorship: the machine is questioned, extracted from, curated, and contradicted. If the paintbrush was a technical revolution, it would be sterile to use it only to paint other paintbrushes. Lingering at the edge of the dreamlike images these systems produce, without a decisive human intervention, means stopping the artistic gesture at the threshold of computational intelligence without crossing it. 

Which returns us to the epigraph. I could continue. You chose to stop. The machine’s account of the situation is exactly right, and it is also the entire case for the human position within it. The machine produces possibilities without understanding them. The human decides which possibilities deserve to exist. 

Augmented imagination 

The book’s title stages a substitution that is also its thesis: from AI as artificial intelligence to AI as augmented imagination. 

By augmented imagination we mean a mode of cognition in which these systems function neither as autonomous minds nor as subordinate instruments, but as an extension of aesthetic thought — a shared cognitive zone where human intentionality and machinic generativity intertwine closely enough that authorship becomes distributed. Computational art, on this account, is not defined by the use of algorithms. It is defined by the emergence of indeterminate authorship as an operative condition: a creativity that exceeds both agents and appears as a third, co-emergent intelligence. 

The condition has a recognisable practical signature, and every artist working this way will know it. A system instructed toward one thing produces an anomaly — an impossible anatomy, an implausible spatial logic, a structural inconsistency. The decision that constitutes the practice is what happens next. Correct the anomaly and you have used a tool badly. Recognise conceptual value in an outcome you did not seek, and let it govern the decisions that follow, and you are somewhere else: in what we call the heterogenesis of ends, where outcomes emerge as systemic and unforeseen consequences of intentional decisions, and the work stops being a project executed to plan and becomes an event. 

Grifola Frondosa was made this way, and then read this way, and the two are the same operation performed at different scales. The image itself came out of feedback loops between human and non-human intelligences working in the dark, inside what the book calls another latent space, where code and image and symbol meet. Nothing in it was placed: the mycelial symbolism, the hint of nuclear contamination, the radiographic light — each arrived by stratification rather than by design. Prompting, in that process, was not a technical procedure but something we would defend calling a poetic one: a language suspended between code and invocation, refined the way one refines a hermetic poem, on the understanding that the artistic gesture is not in the click but in the tension toward the indeterminate. 

The five critics simply continued that stratification with other means. What the book finally proposes is not that machines can make art, nor that they can judge it, but that both making and judging have become distributed operations — and that the human position within them is not authorship in the old proprietary sense, but the capacity to decide what must not be produced, even when production is possible. 

We do not make art to speak about the machine. We make art in spite of it. 

Andrea Filippi and Gabriele Moschin work together as Filippi Moschin, at the intersection of philosophy of technology, aesthetics, and artistic practice. 

Grifola Frondosa. From Artificial Intelligence to Augmented Imagination (2025) — https://doi.org/10.5281/zenodo.21721009 

AI Art Catalogue 2023–2025  — https://doi.org/10.5281/zenodo.21676739