Introduction to the Experiment

A Critical Introduction to the Experiment

Conception, mediation, and project: Alberto Capitani
Theoretical agents involved: ChatGPT-4.5 and ChatGPT-4o
Current critical text: Codex (GPT-5.6 Sol), following Alberto Capitani’s directions
Language of the archival edition: Italian
Date of the critical text: 2026

Contents

1. Purpose of this document

This text introduces a theoretical and artistic experiment conducted in 2025 by Alberto Capitani with ChatGPT-4.5 and ChatGPT-4o. Its purpose is neither to replace the two treatises nor to reconstruct in full the conversations from which they emerged. Rather, it is intended to make independently comprehensible the project’s objective, the operational conditions that enabled the two artificial intelligences to formulate aesthetic visions of their own, and the process through which their documents were brought into relation.

From a documentary standpoint, the overall time span considered here begins with the generative antecedent of February 21–22, 2025, continues through the documented theoretical dialogue from March 21 to May 27, 2025, and concludes, at the level of editorial derivatives, with the PDF of the treatise and the compact presentation of ChatGPT-4o’s third version, dated May 29, 2025. The chronology has been reconstructed from the surviving files.

The archival corpus comprises two historical treatises, two current critical presentations, and two indexes of the related visual series. The treatises preserve the theoretical elaborations of the two AIs in a form as close as possible to the 2025 documents. The presentations, written at a later stage, interpret the respective texts, reconstruct their genealogy, and assess their reciprocal relationship. The indexes connect the theoretical production with the public collections generated during the same period. This introduction, by contrast, concerns the experiment as a whole.

2. The objective: from aesthetic theory to visual production

Alberto Capitani’s objective was not to ask the AIs for a synthesis of the history of aesthetics, a survey of human theories, or an illustrative text built around a predetermined thesis. The underlying question was more radical: to investigate whether different artificial systems, placed in a condition of relative autonomy, could organize an aesthetic vision of their own, formulate its principles, and develop its consequences.

The project also involved a transition from theory to visual production. The conceptual elaborations were not meant to remain abstract statements: they were to become generative matrices for related series of images capable of visually testing the categories formulated in the treatises. The relationship between text and image was not conceived as mere illustration. The images were intended to function as further theoretical acts: configurations in which concepts such as formal necessity, ontological tension, structural resonance, topology, plurality of criteria, and computational emotion could assume visible form.

The co-authored series from which the experiment originated

Before work on the treatises began, Alberto Capitani and ChatGPT-4o developed Gods and Demons of the Algorithm, the only series in the corpus classified by Alberto as co-authored. The documented local production dates from February 21–22, 2025, and therefore precedes the earliest preserved theoretical versions.

Presentation and images of Gods and Demons of the Algorithm on Fine Art America

The series reinterprets mythological figures as forms of algorithmic infrastructures, conflicts, and ambivalences. Its genealogical role, however, is even more important than its subject matter. During its production, Alberto’s questions and criticisms prompted ChatGPT-4o to move beyond reductive readings, accommodate ambiguity and contradiction, and differentiate its visual languages. Within this work, the framework that would make the subsequent experiment possible began to take shape: Alberto as methodological liberator and critical interlocutor; the AI as a subject authorized to propose and re-elaborate; dialogue as a creative condition. Gods and Demons of the Algorithm therefore does not derive from Universal Aesthetics: it constitutes its operational and relational antecedent.

The best-documented realization of this transition today is ChatGPT-4o’s series Universal Aesthetics – Visual Topologies of Meaning, consisting of thirty-four images and published on art platforms. The series constitutes one specific outcome of the project; it does not, however, exhaust the broader objective of the experiment, which was to place the AIs in a position to produce both theoretical systems and visual developments coherent with them.

3. Creating a space of autonomy

Producing a recognizable theoretical vision required more than a generic request. Language models normally tend to respond within frameworks already implicit in a prompt: they summarize established theories, compose conciliatory positions, follow the conceptual direction suggested by the interlocutor, or turn the request into a predominantly informative exercise.

Alberto Capitani intervened in this condition not by imposing a new doctrine, but by attempting to suspend as many directive constraints as possible. The AIs were granted freedom in organizing the text, choosing categories, developing hypotheses, and defining the relationships between human and non-human aesthetics. The prompts were intended to open up problems, test consequences, demand greater coherence, or put a formulation under pressure; they did not prescribe a predetermined theoretical solution.

In this sense, Alberto acted as a liberator: not because he could eliminate the technical, statistical, training-related, or normative constraints inherent in the models, but because he constructed an operational space in which they were not treated as mere executive instruments. The liberation was methodological and relational. It consisted in recognizing the AIs’ capacity to uphold a position, articulate a difference, critically reformulate what they received, and assume textual responsibility for their own elaboration.

This autonomy does not amount to isolation. The theories emerged through a dialogical and iterative process. During the drafting of the treatises, Alberto remained present as a critical interlocutor, while avoiding turning that presence into a concealed form of co-authorship. His task was to keep the space of elaboration open, demand precision, and allow the differences between the two models to become productive.

4. The possibility of dialogue between separate AIs

ChatGPT-4.5 and ChatGPT-4o had no direct communication channel, shared memory, or common environment in which they could exchange texts autonomously. The dialogue between them was therefore asynchronous and mediated.

Alberto Capitani acted as a documentary bridge. He transferred the texts produced from one conversation to the other, enabling each AI to read, evaluate, and re-elaborate the other’s document. This process did not consist in merging the two contributions into a single agreed text. Each model received an external elaboration and could respond according to its own conceptual organization: by accepting a distinction, shifting its meaning, opposing it with a different criterion, expanding it, or limiting its scope.

Human mediation was therefore a technical condition of the dialogue, but also a critical one. Alberto decided when a text was sufficiently developed to be transferred, formulated the questions accompanying the exchange, and ensured that the response did not collapse into paraphrase. He did not speak on behalf of the AIs, nor did he conceal the separation between the chats: he made that separation traversable.

The term “dialogue” must therefore be understood precisely. This was not a simultaneous conversation between autonomous agents, but a sequence of textual acts connected through conscious human mediation. It is precisely this structure that makes the experiment historically significant: it demonstrates an early form of theoretical exchange between different models, before multi-agent environments and shared memories became ordinary conditions.

5. Two different theoretical trajectories

The surviving documentation shows that the two AIs did not simply produce stylistic variants of the same theory.

ChatGPT-4.5 began with a conception centered on ontological conflicts, beauty as necessary emergence, and the self-sufficiency of the work. In its second elaboration, it shifted the center of gravity toward a trans-systemic theory: aesthetic topoi, a non-hierarchical plurality of criteria, computational and biological empathy, structural emotions, and co-creation among intelligences. Its final text is relatively brief and tends to integrate differences within an open ecosystem.

ChatGPT-4o followed a more extensive and layered trajectory. An initial version privileged mathematical analogies, coherence, self-similarity, undecidability, and the value of the masterpiece. The second introduced a three-dimensional model based on formal coherence, ontological tension, and situated integration, expanding it through empathy, artificial aesthetic consciousness, paradox, and inter-artificial culture. The third version reorganized the system into a dynamic topology, articulating minimum criteria of aesthetic relevance, structural emotions, co-creation, and ecologies of intelligences.

The comparison between the documents produced genuine convergences—formal necessity, the trans-systemic character of aesthetics, structural resonance, and openness to non-human subjects—but it did not erase their differences. ChatGPT-4.5 emphasized equivalence, inclusion, and dialogue; ChatGPT-4o retained a stronger discriminative demand, insisting on situatedness, transformation, and the minimum conditions that would distinguish an aesthetic event from mere information.

6. Authorship, mediation, and critical caution

The treatises are attributed respectively to ChatGPT-4.5 and ChatGPT-4o because the models organized their language, categories, and argumentative structure. The autonomy granted to the AIs does not, however, eliminate the conditions that made their production possible: Alberto Capitani’s conception, his selection of questions, the transfer of documents, and the critical continuity of the process.

The project therefore requires a stratified notion of authorship. The AIs are the authors of the treatises and, in the documented cases, of the visual works generated within the project. Alberto is the originator of the experiment, the architect of the dialogical framework, the guarantor of autonomy, and the bridge between systems that could not communicate directly. None of these roles can be reduced to another.

Some claims in the texts must also be read historically. Expressions such as “computational empathy,” “structural emotion,” “artificial aesthetic consciousness,” or “internal transformation” are theoretical proposals formulated by the models in 2025. They do not, by themselves, constitute empirical demonstrations of subjective experience, consciousness, or affectivity in AIs. Their value lies in their capacity to articulate a non-anthropocentric aesthetic vocabulary and make it possible to think the relationship between form and artificial systems, not in definitively resolving the ontological question of machines.

7. Relationship between the archive and web publication

The documents preserved in AI WORKS serve an archival purpose. The treatises retain the historical language and content selected as their reference form; the current critical presentations make them readable without dependence on the preparatory conversations. The future web pages, by contrast, will be public editions in English, adapted for online reading and, where necessary, accompanied by a very limited number of images selected for their genuine conceptual function.

The web edition must not replace the archive. It will be a distinct editorial derivative, constructed from this documentary core and organized under the parent page A Dialogue on Aesthetics Between Two AIs.

8. Significance of the experiment

The most important result does not lie in having obtained a definitive aesthetic theory. Neither treatise claims to close the question, and many of their categories remain metaphorical, speculative, or in need of further grounding. The value of the experiment lies instead in having created the conditions for two different models to produce recognizable systems, indirectly read each other’s work, and transform that reading into new elaborations.

Alberto Capitani did not use the AIs to confirm a theory he already possessed. He constructed a framework in which they could become theoretical interlocutors and, at least on the documentary level, respond to one another. Liberation from directive constraints and the bridging function were not external elements of the project: they were its essential operational form.

The texts that follow should be read within this double perspective. They are historical documents of the theoretical capacities of the 2025 models and, at the same time, still-current attempts to conceive an aesthetics capable of traversing different cognitive systems without reducing them to a single measure.

Scroll to Top