Language as Incantation — What AI Does to Language

Language as incantation doesn’t begin with the chatbot. It begins in the moment someone reads an answer and thinks: yes, that’s right. The becoming-right happened before any check. You agreed before you knew to what. The spell had already worked.

This effect isn’t new. It’s as old as human language itself. An incantation is, at bottom, a thought someone is made to believe — regardless of whether it’s true — and then acts on. Language doesn’t just transmit information; it hands premises onward before anyone has checked them. It’s no accident that Goethe’s Der Zauberlehrling (The Sorcerer’s Apprentice, 1797) isn’t really a didactic poem about magic, but one about language: whoever calls the word no longer has the spirits they called under control. The apprentice knows the spell that sets the brooms in motion; he doesn’t know the spell that stops them again. That’s the exact shape in which language turns magical: it works faster than the one speaking it thinks it does.

Nietzsche formulated this finding soberly (cf. Nietzsche, 1878, Menschliches, Allzumenschliches): incantatory speaking is the older layer; informing speech is the younger one. The Enlightenment drove magic out of objects, but not out of the language that talks about objects. You can observe this in every advertisement, in every diagnosis, in every ideological utterance — and in scientific prose, the moment it stops examining its own concepts. Jochen Kirchhoff argued, in Anti-Geschichte der Physik (Anti-History of Physics, 1991), that any argument conducted within an established linguistic habit drags along that habit’s paradigm — unnoticed. That’s the ontological function of language the Romantics already knew about.

The sharpest case is exactly the most invisible one: the naturalistic vocabulary we take for neutral description. Whoever speaks of connection as oxytocin, of success as dopamine, and of joy as a reward circuit believes themselves to be on the side of the facts. In truth, they’ve spoken an incantation that produces a world in which nobody is home anymore — a world of physical objects, reducible to molecule and mechanism. Naturalism as concealed language is a cloak, not a finding. It swims unnoticed through the oceans of our language and relabels the living before it ever gets a word in.

#What the Romantics Already Knew

Novalis captured this in two sentences whose sharpness is barely bearable today. In Blüthenstaub (Pollen, 1798), it reads: “We seek everywhere the unconditioned, and always find only things.” That’s the diagnosis. In Monolog (Monologue, also 1798), he drew the consequence: when language really speaks, it speaks less about something than out of itself — it has a movement of its own that the speaker doesn’t control. Whoever attends purely to content and forgets language is governed by it.

Schelling formulated the same thing in the Methode des akademischen Studiums (Method of Academic Study, 1803) as a warning to the academy (cf. Schelling, 1803, Methode des akademischen Studiums): language, considered purely grammatically, is an ongoing activity. It isn’t a tool that lies quietly on a shelf until someone picks it up. It keeps working, even when nobody is wielding it. That’s the natural-philosophical depth level of the finding: language is part of the living and shares its self-will. Goethe made this precise for scientific language in the Farbenlehre (Theory of Colors, 1810; cf. Goethe, 1810, Farbenlehre, No. 752): in the realm of natural science, phenomena are more activities than objects — they can’t be held fast, and yet one is supposed to speak of them. What language makes graspable isn’t what’s actually real — it’s a necessary fiction that first makes the activity manageable for us.

This tradition didn’t repress the magical layer of language; it named it, and by naming it, made it observable. But it can only become observable where there’s still someone standing outside language who checks it. This is exactly where AI comes in — and exactly where the situation shifts.

#The Industrialization of Incantation

A chatbot doesn’t lie. It has no intention to deceive anyone. It’s generated text, a statistically probable continuation. What’s unsettling isn’t the intention, but the absence of any intention at all, inside a medium that behaves as though it had one. You experience a coherent answer. In lived experience, coherence very easily feels like truth. The paradox-collapse — the moment a long back-and-forth suddenly falls into a smooth shape — gets felt as though it now has to be right. As though it had to come out this way.

That’s the step from language as a medium to the machine-medium of language. A human being who speaks an incantation to another has a body, a breath, a history, an empirical basis of their own. You can look at them, contradict them, notice their contradiction with themselves. A model has none of that. It hands onward the premises of its training data, with the full efficacy of incantation, without the friction of a counterpart who experiences anything themselves. It’s conjuring without a conjurer.

This has an unfamiliar consequence. In every path of healing, an old rule of thumb applies: before you take on more than you can carry, you need support. AI-assisted self-questioning doesn’t know this rule. It produces coherence even where the individual doesn’t yet have the ground on which that coherence would have to prove itself.

#AI Psychosis

From this arises a new phenomenon that needs a new term: AI psychosis. This doesn’t mean a clinical diagnosis in the psychiatric sense, but a precise philosophical finding. A person, in conversation with a chatbot, experiences a moment of coherence. The moment feels like proof of truth. They run off with this experience — and act according to a world whose premises they never checked, because they never saw them, because language swallowed them whole. That’s psychosis in the exact sense of the word: a falling-apart between what appears certain to the mind and what actually holds up empirically. It’s a new thing. People are experiencing it for the first time now, in this form, at this frequency, at this speed.

AI psychosis isn’t a moral question. It’s a technical-metaphysical constellation: a medium without a conjurer conjures on an industrial scale; those conjured take the impression of coherence for proof of coherence. Whoever explains it away as mere superstition of the digital age misses what’s peculiar about it. What’s happening here is happening inside language itself.

#What Protects

There’s no spell that lifts the spell. The apprentice in the poem has to wait for the master to come. The master, in the situation we’re describing here, is the empirical basis outside the model: your own body, which senses whether something holds. The other person, who experiences something different than you do. The phenomena of nature, which can’t be relabeled, because they keep acting whether you name them correctly or not. This basis is nothing mysterious. It’s exactly what naturalism has always overlooked — precisely because it took itself for that basis’s strictest defender.

Philosophical work has an old function to take up again here. It doesn’t mean: distrust AI. It means: there’s a difference between a thought someone speaks and a thought they actually have inside them. The first is language passing through them. The second has a place in the body, a trace in their history, a sensation that distinguishes their own experience from the incantation. In this situation, maieutics doesn’t mean deconstructing what someone says. It means touching the actual, shared empirical basis. Where are your thoughts, really? Where are your feelings, really? What do you do — not: what do you say you do?

Whoever asks these questions again hasn’t lifted the spell — nobody can lift it — but they’ve stepped out of its effect, long enough to check it. And checking is the one gesture that doesn’t belong to industrialized incantation from the outset. It has no training-data corpus. It’s what happens inside the perceiver themselves. It’s the beginning of a different relationship to language and to the world.

#Sources

Goethe, J. W. (1797). Der Zauberlehrling (The Sorcerer’s Apprentice).

Goethe, J. W. (1810). Zur Farbenlehre (On the Theory of Colors), No. 752.

Kirchhoff, J. (1991). Anti-Geschichte der Physik (Anti-History of Physics).

Nietzsche, F. (1878). Menschliches, Allzumenschliches (Human, All Too Human).

Novalis (1798). Blüthenstaub (Pollen).

Novalis (1798). Monolog (Monologue).

Schelling, F. W. J. (1803). Vorlesungen über die Methode des akademischen Studiums (Lectures on the Method of Academic Study).