“I went to the gas station and when I was paying the girl at the register said:
‘You smell good’.
You can’t have this interaction with ChatGPT.”
— Ricardo Araújo Pereira (Expresso)

When you hear someone like Ricardo Araújo Pereira state, with accurate irony, that “you can't have this interaction with ChatGPT”, the phrase echoes immediately and intuitively: human, physical, sensorial and unpredictable experience seems to be a protected territory, impervious to the action of a machine. It's a seductive statement, because it reassures us. However, seduction is also a risk: it lulls us into false clarity and helps to fuel the mystification surrounding artificial intelligence, both in the sense of its future omnipotence and its supposed present impotence.

What is missing from this reasoning is the awareness that everything current AIs produce is a consequence of human networks of decision, intention and interpretation. There is no autonomous subject acting here. When a user writes a question, when a programmer defines a parameter, when a team decides what data goes into training, the machine just executes, recombines and returns results that will be read and experienced by humans again. Even systems for automatically creating and disseminating fake news, which operate without direct manual intervention, exist because someone designed them for that purpose, with specific objectives. The “autonomy” here is purely instrumental, a delegation of tasks, not a will.

This distinction seems obvious, but our contemporary culture has an almost instinctive ability to forget the genealogy of its own artifacts. The worst thing a human can do is ignore their own achievements, whether admirable or dangerous. Because when we forget the human authorship of machines, we give in to the temptation of treating them as “other” entities, or as inevitable demons, or as neutral oracles. We lose the possibility of taking responsibility for their existence and the way we use them.

There is an irony here: the same public that daily consumes content written by absent humans, chronicles, novels, comedy scripts, social media posts, podcasts, without any physical contact with their authors, tends to accept without question the idea that the experience mediated by AI is, by definition, qualitatively inferior to the experience mediated by humans. Now, textual mediation is always, in itself, an amputation of the body. Even when a writer writes to provoke a laugh, a shiver or a cry, he trusts that the words, in their “disembodied” state, will have an effect.

If we want to understand how far we can go with artificial intelligence, the first step is to recognize that it did not fall from the sky, nor did it sprout from nowhere. It is an extension of the human ability to design, structure and delegate thought. Ignorance, in this field, is not just a lack of information, it is a dangerous stance, because it transfers responsibility for processes that entirely belong to us to the machine.

Criticizing AI is necessary; Ridiculing it can also be healthy, as an antidote to the triumphalist rhetoric of Silicon Valley. But this criticism only gains strength if it is made with lucidity about its real place in the chain of cultural and political production. Otherwise, humor becomes a veil that covers the central issue: it is not the machine that decides what to say or do; We are the ones who decide, even if sometimes we decide badly.

The challenge, therefore, is not just measuring “what AI can or cannot do”, but asking “what do we want it to do” and “what degree of delegation are we comfortable with”. Human history is full of tools that expanded the scope of our action, from writing to the steam engine, from the press to cinema, and all of them were accompanied by enthusiasms, resistances and perverse uses. The difference is that, in the case of AI, the speed and opacity of the processes make forgetting human authorship much easier.

If we allow this forgetfulness to take hold, we open the way to two complementary traps: that of fatalism (“AI will take over everything”) and that of innocence (“AI is just a tool, it has no political importance”). Neither position serves to guide sound ethical and political decisions. What we need is to keep in mind that these technologies are products of human intentions and that, as such, we can criticize, reformulate, restrict or even dismantle them. But to do so we have to understand them, not as an absolute “other”, but as the mirror, distorted and amplified, of our own action.

Ignoring this mirror, whether to venerate it or to demonize it, is to blind ourselves to the responsibility we have. It is giving up the right to decide how far we want to go. And, like every time we abdicate, someone will decide for us.

The question of authorship and responsibility becomes even more interesting when we place it in the editorial field. The proposal of publishing futures, as a critical, speculative and transition design practice, offers precisely a platform to try to deal with this problem without falling into the traps of naivety or technological panic.

Ultimately, editing is not just gathering content and placing it on a support; it is also to inscribe in time and space a specific relationship between those who produce, those who distribute and those who receive. In the case of artificial intelligence, assuming this editorial logic means recognizing that each text generated, each image created, is the result of a complex network of human actions: from the programmer who wrote the model's architecture, to the curator of the training data, to the user who formulated the request, to the reader or spectator who attributes meaning to the final work.

If we think of publishing futures as a testing ground, then AI becomes raw material for investigating not only content, but the chain of mediation itself. It is possible to imagine publications that explicitly reveal the genealogy of each text generated: the prompts used, the relevant training sources, the conscious choices and omissions of the human editor who frames it. Not to “expose” the machine, but to reinscribe its production within the territory of shared responsibility.

This gesture has political implications. When a reader receives a text from a magazine, newspaper or book, they are rarely invited to learn about the material and relational process that generated it. With AI, this opacity increases: the scale and complexity of processing create an even greater distance between source and result. By bringing to the editorial field the practice of making this distance explicit, of mapping it, we open space for a new type of critical literacy.

Here, the role of the editor is close to that of an anthropologist who, like Bruno Latour in We Have Never Been Modern, refuses to separate technical facts from social facts. An AI-generated text is not “just” technology, any more than a printed book in the 15th century was “just” paper and ink. It is a hybrid of forces, interests, intentions and accidents and, as such, can only be understood if we look at it as the product of a network.

In the scenario of publishing futures, this network does not need to be disguised or aestheticized as an artificial “intelligence” fetish. It can be presented as an integral part of the work. A book may contain, in its own body, the instructions that led to its manufacture. An exhibition can showcase not only the content generated, but also the human interactions that made it possible. A magazine may include, alongside a text, the family tree of decisions and inputs that shaped it.

This type of editorial practice can prevent AI from being naturalized as an “autonomous entity” or demonized as an “uncontrollable threat”. Reinscribes it in the human field, where criticism, ethics and imagination have agency. And, perhaps most importantly, it invites us to see technology not as an inevitable destiny, but as a field of shared experimentation.

If the worst thing we can do is ignore our own achievements, then publishing futures can be understood as the art of keeping active memory of these achievements, even, and above all, when they present themselves in the form of machines that seem to talk to themselves.

Let’s imagine an experimental project called “Genealogy of a Text”. Each publication would contain two inseparable layers: the final content (text, image, audio) and the explicit record of its production process. In the case of AI-generated text, this genealogy could include: the original prompt, as written by the user; A brief note of context: what motivated this request, what human circumstances are at its origin; A summary description of the model used, including training date and any known limitations; Editorial changes made after automatic generation; A final reflection from the editor on the expected impact of the text and what was left out.

Another example: “Atlas of Interactions”, a publication that collects interactions between humans and AI in a diary format. Each page would document not only the output generated, but also the verbal exchanges, hesitations, reformulations and even physical or emotional reactions that occurred during the process. It would be a kind of editorial ethnography, where the human and the machine are co-participants in an open narrative.

In an expository context, this logic could take shape in the form of editorial installations. Imagine a space where the visitor reads a printed poem, but next to it he sees the history of conversations that led to that result projected, with marginal comments from those who intervened in the process. Or a wall with diagrams showing how certain terms and styles were chosen or discarded. This would not be to “demystify” AI in the sense of taking away its magic, but to shift the magic from the myth of autonomy to the reality of collaboration, even if that collaboration is asymmetric and mediated.

This type of editorial practice fulfills two functions: the first is to recover human authorship as a structuring presence, even when the product seems to “come from the machine”. The second is to offer the public critical tools to understand that every object generated by AI is a cultural, social and technical hybrid, and that its reading requires awareness of this hybridity.

Ultimately, it is about reversing the naturalization process that accompanies any technology when it becomes invisible. AI, like electricity or letterpress printing, tends to become “transparent” in its operation. A critical publishing futures can counter this trend, keeping the seam between human and machine visible.

transition design is not limited to designing objects or systems; it is a field that operates on ecosystems of practices, imaginaries and infrastructures. In this sense, the relationship with artificial intelligence cannot be treated as a mere matter of efficiency or style. It is, first and foremost, a question of active memory and collective agency.

Each technology carries with it the mark of its time: the expectations, tensions, vices and aspirations of those who create it. By ignoring this mark, we not only give up the possibility of criticism, but we also compromise the ability to reconfigure its direction. This is where the proposal of publishing futures intersects with the challenge of AI: creating practices that keep in view the web of human relationships that give rise to content, instead of presenting them as hermetic and self-justifying products.

The editorial gesture, in this context, becomes a political gesture. By showing the genealogy of a text, by giving visibility to the decisions and conditions that shaped it, we return to the public a form of interpretative sovereignty. It is not a question of “explaining” the machine like someone dismantling a trick, but of reinscribing the generated object in a context that allows the reader to assume a critical position.

This practice moves us from a technological imaginary centered on the fascination with the autonomy of machines to another, more demanding one, which recognizes the structuring role of collaboration, even if this collaboration is indirect, mediated and unequal. In transition design, this awareness is essential: each editorial decision can function as a prototype of a future relationship between human and machine, between author and reader, between tool and community.

In this way, AI can no longer be thought of as a threat or miracle, but rather be seen as yet another episode in the long history of human mediations. And, as happened with writing, the press or photography, the way we integrate it will depend less on its “nature” and more on the choices we make around it. Choices that, to be responsible, require us to reject ignorance as a method and adopt cartography as a practice: mapping where each gesture, each phrase, each image comes from, to decide clearly where we want it to go.

Maintaining active memory of our own artifacts is more than a historical exercise; it is a way of ensuring that the future is not built at the expense of erasing our intentions. In publishing futures, this translates into publications that function not only as objects of reading, but as devices of revelation, where aesthetic experience and critical consciousness meet.

An employee's spontaneous compliment at a gas station may, in fact, be beyond the reach of an AI model. But the ability to create and mediate speeches that provoke attention, reflection and even affection, even without a body present, is not exclusive to humans. It is a shared territory, which requires us to be lucid so as not to confuse authorship with the interface, nor myth with practice. And above all, not to forget that, as long as humans program, edit and interpret, AI will continue to be a mirror, not an oracle.

Appendix: Genealogy of this Text

Final reflection
– The text is the result of a prolonged and deliberate collaboration between human and machine, where the user maintains authorial, conceptual and editorial control.
– The genealogy described here serves not only to record the origin of the text, but also to reveal the process as an integral part of the work — in coherence with the thesis defended throughout the essay: the importance of reinscribing human authorship in everything that is mediated by AI.

Initial request
– The starting point was Ricardo Araújo Pereira's phrase, used as a motto to dismantle a popular idea about the supposed inability of AI to create experiences comparable to human ones.
– The user immediately introduced the central thesis: “The worst thing a human can do is ignore their own achievements.”

Human and historical context
– This text is part of a journey of several months of continuous work in which the user has been uploading bibliographic information, sharing references, exploring concepts and developing ideas linked to publishing futures, transition design and cultural and technological criticism.
– Throughout this period, the user uses the paid version of ChatGPT (GPT-5), ensuring greater continuity, context capacity and access to more extensive resources.
– Dialogue is not one-sided: there are frequent questions, criticisms, corrections and reformulations requested by the user, which guide the tone, structure and density of the text.
– The work is not limited to automatic responses: it involves hesitation, doubt, trial and error, with successive revisions and deepening, rarely stopping at the first writing.

Model and production conditions
– GPT-5 from OpenAI, configured to write in Portuguese from Portugal, with an affirmative tone, formal rigor, organic integration of references and absence of negative constructions.
– The interaction took place in an iterative mode, with the user requesting expansions, adjustments to the conceptual order, and integrations of pre-existing excerpts in their personal file.

Process
– Layered structure:

  1. Criticism of the RAP joke and exposure of the problem of authorship and the mystification of AI.
  1. Connection to the editorial field and the concept of publishing futures.
  1. Presentation of practical proposals (Genealogy of a Text, Atlas of Interactions, editorial installations).
  1. Integration of the chapter on transition design, active memory and collective agency.

Closure with repositioning of AI as a mirror and not an oracle.
– Each layer has been revised and consolidated from previous versions, incorporating both detail adjustments and rewrites of entire sections.

Editorial intervention
– Syntax and vocabulary adjustments to preserve consistency and argumentative density.
– Integration of bibliographic and authorial references previously worked on in other conversations (Latour, among others).
– Review to ensure narrative fluidity and coherence between the parts.