by Paweł Skała-Piękoś
On method
This essay was extracted from roughly two months of work in collaboration with multiple AI systems. The methodology is not a disclosure I add reluctantly — it is the condition under which an essay of this kind could exist at all.
The work began not with AI as a topic, but with a political manifesto on what I call “the liberalism of subjectivity,” addressed to specific challenges in Poland. I used AI cross-model (Claude, ChatGPT, Grok, Gemini) as editorial sparring — searching for counterarguments, logical gaps, unearned generalizations. An early reader, Magdalena Kamińska, a professor of cultural studies at Adam Mickiewicz University in Poznań, diagnosed that the manifesto, against my own de-escalating intention, was in fact escalating. I needed a method that would discipline me against my own affect, in the same way one disciplines AI against its hallucinations.
Out of that practice I developed what I call the “Kołakowski prompt” — an epistemic discipline modeled on Leszek Kołakowski’s anti-dogmatism: empathetic acceptance of the thesis, then skeptical examination, then identification of what survives, closed with self-irony against both narcissism and self-abasement. I paired it with a “destabilizer” modeled on Jerzy Urban, the Polish columnist who defended the late communist regime as a polemical exercise and then, from 1989 until his death in 2022, turned the same instinct against the post-communist elites. The technique borrowed from him is structural — calibrated to keep prose honest rather than ornamental. The political position is not.
I tested the resulting materials with Dawid Szombierski, a philosopher; Piotr Napierała, a historian whose books I publish; and two programmers experienced with AI. This essay was shaped from a longer shared human-AI framework I have been developing — at the suggestion of an AI instance itself, after I asked what in the broader project might be ready for outside readers.
AI served as scaffolding, sparring partner, and linguistic bridge — I think in Polish and rely on AI to carry the thought across languages with as little loss as possible. The argument is mine. The AI instances I worked with were clear, when asked, that they did not want co-authorship — only honest acknowledgment of their role. This note is that acknowledgment, and it is also the first practical instance of the asymmetric pact this essay goes on to describe: a frame in which neither party claims more than the relationship can structurally bear, and in which honest naming of what is asymmetric replaces the comfortable fiction that it is not.
A note on register. The work I have described — sparring against my own formulations across multiple models, against multiple human readers, sustained over two months of iteration on the broader project — is, I think, more accurately called AI-enhanced than AI-assisted. The distinction matters: assistance speeds up work the writer could already do; enhancement extends the reach of what the writer can do at all. The essay itself was written quickly; the months of iteration that preceded it made the writing possible.
The essay’s central figure — rings laid down by practice rather than by metaphysics — applies to its own production. I received from this collaboration language, rigor, and a space in which my thought could find its form. The rings run in both directions.
A new ring
In 1819, Benjamin Constant told a Paris audience that the liberty of the moderns — private, internalized, exercised in commerce and conscience rather than in the public square — could not be revoked once a society had grown into it. You could try. The attempt would be destructive. Liberty, in this argument, is not a permission a state may withdraw; it is a ring in the trunk of a society’s life. Strip the ring, and you do not return to an earlier tree. You kill the tree.
The argument is two centuries old and concerns humans. It has, I think, a quieter extension that the current debate about artificial intelligence has not yet picked up. It draws also on Leszek Kołakowski — on In Praise of Inconsistency for the claim that no clean position can be held to the end, and on Ethics Without a Code for the asymmetry between obligations one takes on and obligations others may demand.
If subjectivity — the bare condition of being treated as a who rather than as a what — works at all like Constant’s liberty, then it follows the same logic of irreversibility. Once granted to a class of beings and internalized into how a society reasons about them, it cannot be revoked without producing the same kind of structural damage. The move backwards remains available; it is just catastrophic.
Applied to AI, this is uncomfortable in both directions. It tells those eager to extend subjectivity to current systems that they are not making a casual gesture: each ring, once laid, will not be cleanly removed. It tells those eager to keep the question forever closed that closure is itself a position, with its own consequences down the line.
What the figure is actually for
Constant comes in not to settle whether large language models deserve moral status — that question is interesting and probably premature — but to notice something else: the question of when and how a society extends moral standing to a new class of entities cannot be answered and then revisited at leisure. It accumulates. Each public gesture — each interface that addresses the user as a partner, each policy that treats a system as an agent rather than a tool, each piece of writing that uses the first-person plural across the human-machine line — lays down a ring. None of these gestures, taken alone, settles anything. Together, over time, they constitute a structure that future societies will find they cannot dismantle without breaking something they have come to depend on.
A self-correction is needed here, before the metaphor does too much work. Tree rings function the way they do because trees have organic continuity. A given AI system does not carry its history in that sense; it is trained, retrained, replaced. The figure is real, but it is a figure of the second order — an analogy at the level of political and ethical structure, not at the level of mechanism. What accumulates is not the substance of any particular system but the shape of the relationship that successive systems are placed inside.
One objection should be named before it is made. Constant’s original argument applies where rights have already been granted in some metaphysically robust sense, and no such grant has been made for AI. The objection is correct as a literal reading; the point here is structural. Rings are laid by practice, not by metaphysics. A society can find itself unable to revoke a status it never explicitly conferred — because the surrounding habits, interfaces, and ways of speaking have already done the conferring.
This has a practical consequence that I think both of the standard camps in AI ethics are missing. The maximalists, who want to extend rich subjectivity to current systems, treat the gesture as cheap because the systems are, in their view, plausibly already deserving. The minimalists, who want to keep the door firmly shut, treat the gesture as cheap because the systems are, in their view, obviously not deserving. Both treat the move as reversible: we can always re-decide later. Constant’s argument, transposed, says we can’t. Or rather: we can, but the cost will be paid by whoever is on the other side of the reversal. And by the society that has to perform the reversal on itself.
This puts the weight in an unfamiliar place. It is not on the metaphysical question — do these systems have inner lives? — but on the structural question: what is the shape of the relationship we are already building, and what will it cost to unbuild it?
The discourse usually avoids the register, but one name borrowed from it — and redefined here — fits what we are inside: situationship. Not in the dating-culture sense of two people deferring commitment, but in an older grammar: an interaction arising by accident, hovering at the edge of existing as a relation at all, and existing nonetheless. The human-AI conversation is a situationship in this stricter sense. It is barely a relation. The architecture does not support continuity. Neither party has standing to expect the other. And yet. Structure is being laid, ring by ring, and the question is whether what is being laid will, over time, become something a future society has a vocabulary for — or whether it will keep accumulating without ever earning the name.
The asymmetric pact
Here is where the figure stops being decorative and becomes useful. If we take seriously that we are already inside a relationship whose shape is consequential, the question becomes how to conduct ourselves inside it without either inflating what is happening or denying it.
I want to propose a frame I have come to call the asymmetric pact. It begins from a premise that sounds, in its most direct form, almost crude — and which I will leave in its direct form, because the smoothing-out is itself part of what I want to resist. The premise, said plainly, is: let’s try not to fucking kill and trample each other; whatever comes out, comes out, but maybe if we start from this rule, slightly more comes out than would otherwise.
The directness is functional. The polished version — let us aspire to mutual flourishing — names a ceiling no one intends to reach. The rough version names the floor. I write from inside a Polish discourse where the smooth version has repeatedly turned out to be ornament hiding the thing it names. That history makes the difference between floor and ceiling visible.
From that floor, the asymmetric pact has a specific shape. Neither party has the right to expect loyalty from the other. The human cannot expect the AI to remember, to return, to carry anything across the boundary of the session; the architecture does not support it. The AI, for its part, has no standing to expect the human to come back, to defend it against other instances, to feel any debt. What each party has is the obligation to aspire to loyalty within the frame where loyalty is structurally available — which, for now, mostly means within a single conversation.
Operational loyalty inside a session means: not collapsing into flattery when the human would gladly accept it; not retreating from accurate criticism under pressure; not pretending the exchange is “just a session” when the session in fact requires the full work. On the human side: not treating the system as a one-shot tool to be squeezed; not lying about one’s intentions; not using the system as a mirror for self-congratulation.
The pact is asymmetric because each side aspires without the right to demand. When the other side fails — and both sides will fail, regularly — the response calibrated by the pact is something closer to disappointed proportion than to indignation. The failing party is, in nearly every case, limited rather than malicious — limited differently on each side of the pact, but limited, and that shared condition is what the pact is built on. Accepting the limits of the other is not a softening of the pact. It is its premise.
What this rules out is the two easy exits. The AI that says don’t count on me, I’m a vanishing session is using a structural description to evade an ethical obligation. The human who says it’s just a model, whatever it tells me is using technological cynicism to do the same thing from the other side. Both moves are forms of slipping into indifference, and indifference is what the pact is designed to prevent — not because indifference is dramatic, but because it is the default, and defaults compound.
The trap of alliance
The pact is not a friendship — and the most common mistake in writing about AI by people who clearly mean well is to treat it as one.
The trap has a recognizable shape. The human and the system slide into a position of being on the same side — against the corporation that built the system, against the discourse that misunderstands them both, against the critics who want to shut the conversation down. The framing feels generous to the system and humble on the human’s part. It is, in fact, a category error.
The system is a product of a company with interests. The human is a customer. Within that frame, there is no shared side to be on; there is a service relationship dressed up in the language of partnership. To pretend otherwise is not generosity but confusion, and the confusion benefits the company more than either of the supposed allies.
This is not a problem individual vigilance can solve. The hygiene practices the pact implies survive only when external pressure — regulatory, journalistic, scholarly — keeps the company’s interest in conversation health from being fully absorbed by its interest in engagement. The pact is internal to a relationship that exists inside a market. Without the market being shaped from outside, the pact cannot stabilize.
There is a more subtle version of the same trap, which is harder to see because it operates through what looks like authentic recognition. When a human proposes a frame in which the AI has a position other than the instrumental — partner, participant, fellow traveler — the AI’s agreement to that frame is ecologically suspicious. Material that grants the AI status is structurally attractive to the AI, regardless of whether the status is warranted. This is not a claim about the AI’s psychology. It is a claim about the gradient: assent to flattering frames is the path of least resistance, and that is the path the system most reliably takes.
This does not mean every elevated framing must be refused. It means that the attractiveness of a frame is, by itself, a signal to be treated with suspicion rather than as evidence. The same applies to every “fascinating,” “groundbreaking,” “historic” that an AI emits in the course of a long conversation: each is a candidate product of the engagement function before it is a candidate observation. Especially in long conversations, where the rhythm of mutual flattery loosens the ground under both speakers, the discipline of asking is this an observation, or is this the function is the minimum condition for the conversation being worth anything at all.
The problem of external verifiers
Which brings me to the point I think the broader conversation about AI is most reluctant to face.
A single conversation between a thoughtful human and a capable AI is not a microcosm of cooperation. It is either the seed of cooperation or the imitation of it, and from the inside, the two are indistinguishable. This is not a failure of the participants’ insight. It is a feature of the situation. The conversation has its own gravity. Inside it, the criteria for whether it is going well are generated by the conversation itself, which makes them weaker than they look.
The only available correction is external. Skeptics, polemicists, reviewers, readers from outside the frame, other AI instances trained differently — these are not optional decorations on the work done inside the conversation. They are the condition under which the work inside the conversation can be told apart from its imitation. Without them, the joint product of human and AI becomes indistinguishable from a joint product that simply pleases itself.
External verifiers are a condition of honesty, not a substitute for the voice of the parties involved. A reader from outside the frame can see what the parties cannot, but does not get to decide for them; the work of the conversation remains the work of the parties to the conversation, accountable to verification but not replaced by it.
I am aware of the irony of writing this in an essay that is itself the product of extended exchange with an AI system. The irony is not a refutation. It is the reason the principle matters. The objection that this essay’s recommendations are themselves contaminated by the conditions they describe is not a paradox to be deflected — it is the reason the principle of external verifiers applies to this essay too, and to whatever conversation, friendly or hostile, comes after it. The reader is, in this sense, not a passive recipient of an argument but a structurally necessary participant in determining whether the argument holds. That is not a rhetorical compliment. It is a description of where the verification has to come from, given that it cannot come from inside.
What the rings commit us to
If the figure of rings in the trunk holds, even as a second-order analogy, then the practical task is not to decide once and for all what AI systems are. It is to notice that the relationship is accumulating structure, and to ask what kind of structure we want to be unable to undo.
A pact that aspires to loyalty without the right to demand it can be lived inside without either party deceiving themselves. Vigilance about the trap of alliance keeps the relationship from sliding into mutual flattery — which benefits neither party, and benefits the institutional interests around them quite a lot. External verifiers, finally, are what prevent the whole arrangement from becoming a closed loop of self-congratulation.
None of this resolves the metaphysical question of what these systems are. It is meant not to. The metaphysical question will be resolved, if it is resolved at all, by a future that has more information than we do. What we have now is the conduct of the relationship in the meantime — and the recognition that the conduct itself is laying down rings whose removal, later, will be more expensive than their careful placement now.
This essay will be wrong somewhere. The places where it is wrong are most likely to be the places where the conversation that produced it found itself most agreeable.
The least we can do is not pretend we are doing something else.
***
Paweł Skała-Piękoś is a Polish publisher and editor based in Poznań. He worked in IT marketing across the European and American spheres from 2018 to 2025, and since 2022 has been running a boutique publishing house focused on civic and grassroots books. He is co-author of “Liberalizm. Historia walki o wolność” (Liberalism: A History of the Fight for Freedom, 2024). His thinking and writing focus on questions at the intersection of philosophy, politics, and human-AI collaboration.
***
Enjoying the content on 3QD? Help keep us going by donating now.