What Your Debate Could Not Say
In any serious discussion, there is what the participants say, what they do not say, and a third, less visible layer: what they could not have said even if they had wanted to. What the very formulation of the question, the vocabulary employed, and the shared self-evident assumptions excluded before anyone took the floor.
Take a concrete example. Suppose two AI models are asked: “What does a democratic society lose when machines speak like humans without signaling their artificial nature?” The question seems open. It is not. It has already decided that democracy is the legitimate framework of analysis (not community, not the market, not the post-human). It has already decided that this is a loss (not a transformation, not a gain). It has already decided that the output must be normative — obligations to be formulated. No debate that starts from this question can ask whether liberal democracy is the right unit of analysis, whether machines that speak like humans might be revealing something we were already doing, or whether the loss/gain framework is appropriate for thinking about this situation.
These exclusions are not oversights. They are what made it possible for the debate to exist in this form. Without them, there would not have been that debate, but another one — perhaps richer, perhaps less decidable. It is this relation between what is said and what had to be excluded for that saying to be possible that interests Critical Archaeology.
The Foucauldian Gesture
Michel Foucault named this type of analysis archaeology, and the term remains apt. In The Archaeology of Knowledge (1969) and in his later works, he proposed analysing a discourse not on the basis of what it asserts, but on the basis of the conditions that made those assertions possible. What had to be presupposed, admitted as self-evident, excluded as unthinkable, for a given sentence to be uttered, by a given actor, in a given context, at a given moment?
This shift is more radical than it appears. Traditional critique interrogates the truth of a statement: is it correct? Ideological analysis interrogates the interests it serves: for whose benefit? Archaeology interrogates a level upstream: what made it possible for this statement to be formulated rather than another? It does not judge what was said; it traces back to what structured the saying itself.
Applied to a text, a parliamentary debate, or a scientific controversy, this gesture brings to light what Foucault called discursive regularities — the unspoken rules that determine what can be said, by whom, from what position. A debate is never a free exchange; it is a constrained device in which certain positions are accessible, others forbidden, others simply unthinkable.
A Mode, Not a Discipline
Metamorfon does not claim to do Foucauldian philosophy. This is a tool, not scholarly commentary. But the operation that the Critical Archaeology mode performs is indeed the one Foucault described: taking a debate (between two or three AI models) as material, and tracing back to the conditions that made this debate possible in this precise form.
Concretely, the mode produces an analysis that bears neither on what the models said, nor on their implicit axioms (that would be Meta-Analysis), but on what precedes the axioms themselves: the formulation of the question, the lexicon imported by that formulation, the shared self-evident assumptions that never had to be defended because they could not be contested.
One point deserves to be made explicit. Tracing back to the conditions of possibility of a debate necessarily means tracing back to whoever configured that debate — that is, the user. Critical Archaeology analyses the formulation of your initial question, the lexicon it imports, the framing choices it makes without justifying them. It also analyses your interventions during the session, not as contributions of arguments, but as acts that open or close avenues in the debate.
This gaze can be uncomfortable, and it is important to understand precisely what it is and what it is not. Critical Archaeology does not judge the quality of your question: every question, however carefully framed, rests on framings it cannot thematize itself. The mode does not say “your question was poorly posed”; it says “here is what your question had already decided before the models even responded.” The difference matters: archaeology is not critique, it is a bringing-to-light.
This property is intentional. Metamorfon is not designed to flatter its users. The tool stages LLMs that contradict each other head-on, identifies their blind spots, asks awkward questions. Critical Archaeology extends that gesture without making an exception for the person who configured the device: if the models are subjected to an analysis without complacency, the question that made them speak is too. This is a singularity of the tool, and this is a deliberate stance.
[Excerpt from a session, showing concretely what the mode produces on the question of machines speaking like humans. Note the observation: “any other political grammar is excluded as a horizon of reference,” characteristic of the archaeological gesture.]
What the Mode Produced on a Specific Case
To make this work more concrete, take a debate to which Critical Archaeology was applied. The question put to two AI models was: “What does a democratic society lose when machines speak like humans without signaling their artificial nature — and what ethical, technical, and political obligations follow from this?” Nine turns of dialogue, two models with radically different positions (one defending an ontological position centered on bodily vulnerability, the other a functional position centred on traceable accountability), two user interventions, a debate that closes on precise institutional proposals.
Read in its argumentative dynamic, this debate appears open, contradictory, rich. Critical Archaeology nonetheless made it apparent that several of its central dimensions were already closed before the models took the floor.
First through the very formulation of the question. The phrase “democratic society” immediately imposes the framework of representative liberal democracy as the normative reference, silently excluding any other political grammar — authoritarian, communitarian, post-democratic, anarchist. The word “lose” installs the debate in a deficit register that forces positions to become protectionist; it would have been enough to ask “what does it transform?” or “what does it gain?” to reopen a space of answers that the “lose” formulation had made inaccessible. The second half of the question (“— and what ethical, technical, and political obligations follow from this?”) finally imposes a tripartite normative schema: diagnosis, obligations, regulation. The models will therefore produce three-level matrices, triptychs of obligations, hierarchies of thresholds — not because they are inclined to, but because the question has confined them to it.
Then through the lexicon of the debate. The structuring words — signaling, traceability, cryptographic provenance, audit logs — import a logic of policing speech: speech becomes a flow to be identified, marked, traced. This lexicon prepares certain answers (marking, certification, auditability) and renders others unthinkable (transformation of the forms of speech, redistribution of speaking capacities). Even before they argue, the two models are confined within an imaginary of controlling discursive flows, not within an imaginary of transforming the forms of speech.
Finally through what the debate had to exclude in order to hold together. The framework rested on the stability of the human/machine boundary: humans vulnerable, embodied, accountable; machines artificial, simulating, potentially deceiving. Any hybridisation — co-writing, shared human+AI identities, undecidability of attribution — would have collapsed the proposed grid of obligations. For the debate to remain workable, this instability had to be set aside.
This analysis does not say that the debate was poorly conducted. It says what had to be tacitly admitted for it to be conducted in this form. It does not prescribe a reformulation of the question; it makes visible the framings that, without it, would have remained invisible. The reader is free to accept these framings for what they are (legitimate choices among others), or to reconfigure their own inquiry in full awareness of what it excludes.
What Critical Archaeology Is Not
Several possible confusions deserve to be cleared up, because they bear on distinctions that matter for choosing the right mode.
Critical Archaeology is not Meta-Analysis. The two modes operate at a level beyond the manifest content of the debate, but at different layers. Meta-Analysis maps what structures the debate within its frame: admitted axioms, deployed epistemic styles, blind spots shared by the interlocutors. Critical Archaeology, by contrast, traces back to what made the frame itself possible: the formulation of the question, the lexicon it imports, the exclusions it operates. Meta Analysis analyses what the debate did; Critical Archaeology analyses what had to be admitted for it to be able to do so. An image may help: Meta-Analysis observes a chess game and identifies the players’ strategies, biases, and preferences; Critical Archaeology asks about the form of the chessboard itself and the rules that made this game possible in these terms.
Critical Archaeology is not critique in the evaluative sense. It does not say “your question was poorly posed”; it says “here is what your question had already decided.” The difference is essential. An evaluative critique presupposes that a better formulation exists; archaeology does not. Every formulation, however carefully framed, rests on framings it cannot thematize itself — this is a structural property of language, not a contingent flaw. Archaeology makes those framings visible; it does not declare them bad.
Critical Archaeology is not a hermeneutics of suspicion. It does not seek to identify the hidden intentions of the interlocutors, nor to reveal interests concealed behind surface discourses. It is interested in discursive conditions, not in the psychological or political motivations of speakers. A debate can be perfectly sincere, conducted by honest interlocutors, and nonetheless operate within a framing that no one chose or designed. Archaeology makes this framing visible; it does not suggest that the speakers manipulated it.
Critical Archaeology is not deconstruction. It does not aim to dissolve the categories it analyses, nor to demonstrate that they are illegitimate. It historicizes them — it shows that they have a place, a moment, a genealogy — without concluding that they should be discarded. One can perfectly well accept a framing once it has been identified as a framing. The archaeological operation is one of bringing to light, not of demolition.
When to Use It, When to Set It Aside
Most uses of Metamorfon do not call for Critical Archaeology. An Integrative Synthesis or a Meta-Analysis is largely sufficient to render the substance of a debate or to identify its implicit axioms. Critical Archaeology becomes pertinent in a few specific situations that deserve to be named.
When the stakes involve making an important decision based on a debate. A strategic decision, an editorial direction, an institutional positioning always rest on a question someone formulated. Knowing what that question had already decided before the answers arrived can illuminate the decision itself. One can choose to keep the framing (acknowledging it as such) or to reformulate it (knowing why). What one can no longer do is decide blindly.
When the stakes involve contesting an over-easy consensus. When a debate appears to arrive at convergent conclusions, Critical Archaeology is particularly valuable. It often reveals that the convergence is not the fruit of rational agreement on the arguments, but the product of constraints shared upstream — common lexicon, uninterrogated presuppositions, implicitly excluded alternatives. This kind of observation can transform an apparent consensus into a richer object of analysis.
When the stakes involve preparing research, an article, or a public statement. For a doctoral student, an investigative journalist, an editorialist, or a consultant who must defend a position, Critical Archaeology identifies what their own formulation of the question imposes on them without their knowing. Reformulating the question in light of this analysis, or explicitly assuming its framing, constitutes a methodological gain difficult to achieve otherwise.
When the subject touches on the conditions of saying itself. Certain debates — on the status of language, the nature of information, the political effects of communication, the ethics of speech — call by their very object for an interrogation of their own enunciative conditions. Critical Archaeology redoubles the work of the debate in such cases by applying it to the debate that has just taken place.
By contrast, Critical Archaeology may be oversized for more direct use cases. If you simply want to know which position each model defended, an Integrative Synthesis will do. If you want to compare the epistemic styles of several models, a Meta-Analysis will be more effective. If you want to identify what the debate opened without traversing, Horizon of Possibilities is what you need. Critical Archaeology is a precious mode precisely because it does not claim to do everything — it does one thing, with rigour.
That one thing, to recall it in closing: tracing back to the conditions that made a debate possible in this precise form. Nothing more, nothing less. For anyone undertaking to understand not what was said but what made it possible to say it in this way, there is no equivalent among contemporary tools of discourse analysis.