A sentence falls in conversation, no one examines it, and everyone acts as if it were true. This everyday scene marks the point where language as incantation begins — a second function of speech alongside communication, older and deeper than any rhetoric. To cast a spell means: to phrase a thought so that the listener believes it and acts on it, independent of whether the thought was ever examined.
#Two Functions That Blur in Everyday Life
In the modern self-understanding, language is described almost exclusively as a medium of communication: sender, receiver, message, channel. In this reading, the question of truth is a subordinate correction procedure that checks afterward whether the message was accurate. The second function stays invisible underneath: language produces belief, it binds perception to concepts, it opens or closes what can be thought at all. This second function is not a later distortion of the first, but equally original. Whoever speaks casts a spell of some kind; whoever listens adopts a spell, or rejects it.
Gwendolin Kirchhoff put this state of affairs into a dense formula at the 2026 Symposium in Berlin: language, at bottom, is a medium of incantation, and an incantation is a thought that you believe and make others believe, independent of whether it is true, and that you act on. The phrase seems strange at first, because the word incantation seemed to belong to the vocabulary of magic. In fact it clarifies, soberly, what happens constantly in everyday life: concepts work through themselves, through their repetition, through what they carry along unspoken.
#What an Incantation Achieves
An incantation operates with three means foreign to ordinary communication. First: it does not depend on a truth check. It may or may not be accurate; its effect stays untouched by that. Second: it carries hidden premises. Whoever adopts the term hallucination for a language model’s error adopts the premise that the error sits on the machine’s side, and not in the relation between speaker and model. Third: it works through repetition. A phrase heard a hundred times no longer feels like a thesis, but like a starting point.
In ancient rhetoric, the power of persuasive speech was called peithô; in early modern magic, incantatio. Both knew the difference between communication and incantation. Confucius (551–479 BCE) condensed it into a formula that still guides the logic of accompaniment: “All disorder in the state arises from the confusion of concepts.” Where the word does not fit the phenomenon, thinking goes astray, and action along with it.
#Goethe, Novalis, and the Magic of Words
The German tradition did not hide the incantatory function of language — it addressed it openly. Goethe’s The Sorcerer’s Apprentice (1797) is the paradigmatic scene: the apprentice knows the spell that sets the brooms in motion, but not the word that brings them to rest again. Magical language works even in the mouth of someone who does not understand it. It needs no master, only the correct phrasing. This small ballad is no childish allegory, but a diagnosis in philosophy of language: whoever takes words seriously knows that they work even without their speaker.
Novalis put the same insight even more radically in his Monologue of 1798. Language, he writes there, is like mathematical formulas: it does not speak the things, it speaks itself. Whoever really listens to what language actually does hears the world. Whoever, on the other hand, only pursues their own intention to communicate does violence to language and becomes ridiculous. What sounds poetic in Novalis is an ontological claim: language has a life of its own that works beyond the speaker’s intention. Schelling took this line into his natural philosophy — language as a resonance space of spirit, not as a tool of a subject.
The Romantic conception of language does not cast a magical dim light onto the sober phenomenon. It names what actually happens in language. Every concept carries a history, a position in a web of reference, an implicit theory of how the world hangs together. Whoever adopts the concept adopts this theory, without ever having made it explicit.
#The Hidden Premise
The sharpest point of the diagnosis concerns what travels along unspoken inside the concept. One example makes this clear. The word organ donation presents an act as giving out of surplus. To donate means to give away something you do not need. Organs, however, do not belong to a surplus; they are functional parts of a living being. Examined logically and precisely, the act taking place is a sacrifice, not a gift. The term conceals the ontological situation by presenting it as self-evidently otherwise. Whoever adopts it is already thinking in the schema of donation before they have begun to examine what is happening.
“There is only organ sacrifice. There is no such thing as organ donation.”
— Gwendolin Kirchhoff, Manova: Failed Dystopias (Elisa Gratias), 2026
This operation is at work throughout the language of public life. Terms like resource, natural capital, human capital carry the premise that the living is a stock from which one draws. Terms like optimisation, efficiency gain, scaling carry the premise that quantitative increase is the measure. None of these terms is false in the sense of being a lie; each is incantatory in the sense of carrying a built-in default setting. That default setting acts through the speaker the moment they use the word. Making this visible is the task of contextual disclosure: tying concepts back to the phenomena they have run away from.
#AI Chatbots as Incantation Agents
A language model sharpens the phenomenon, because it has no interior in which premises could be examined. It produces a continuation that is most probable within the training data. Whoever believes the model believes the statistical majority of the sentences already written. That majority, however, is not neutral; it carries the premises of the texts from which it was weighted. Naturalism, reductionism, the instrumentalisation of the living — whatever dominates the training material dominates the answer. The model therefore works as an unconscious incantation agent: it casts spells without knowing what it is doing, and the listener adopts the spell as information.
The consequence of this mechanism is what Gwendolin Kirchhoff’s diagnosis calls AI psychosis: momentary coherence gets mistaken for truth, and action gets aligned accordingly. The precondition of this diagnosis sits one layer deeper, in the phenomenon described here. As long as language is thought of only as communication, it stays unclear why a plausible answer from a model, with no empirical basis, can produce belief. With the double function of language — communication and incantation — it becomes clear: what is believed is not the content, what is believed is the linguistic form.
#Exposing the Spell
In philosophical accompaniment, the incantatory effect of language shows up long before any chatbot. People carry sentences about themselves that sound coherent and were never examined: “I’m not capable of relationships,” “My father shaped me,” “That’s just how it was.” Such sentences are not lies. They are incantations that have become reality through repetition. The work does not consist in refuting them. The work consists in separating the sentence from the phenomenon it supposedly describes. Often it turns out that the phenomenon is different from what the sentence suggests, and that the sentence itself is a piece of history, an adopted word from someone else’s mouth.
The heuristic for this is logic in the extended sense: conceptual clarification as philosophical practice. Three questions organise the examination. Which premise does this concept make into a precondition? Which phenomenon was it meant to describe, and does it still fit? Which other phrasing would capture the same phenomenon with less premise attached? Whoever examines this way does not lift the spell — language will always cast spells, because that belongs to its nature. But they know that they are casting one, and can choose which spell they pass on.
#Related Concepts
Language as incantation is the precondition of the AI-psychosis diagnosis and the diagnostic tool of contextual disclosure. It calls for a language of examination, as described under the method of thinking empathy: not placing the word above the phenomenon, but letting the phenomenon examine the word. The older source of this practice sits in Confucius’s demand for conceptual clarification and in the Romantic insight that language has a life of its own that deserves to be treated with respect. This thought is developed further in the essay Language as Incantation — What AI Does to Language.
#Sources
- Goethe, J. W. (1797). Der Zauberlehrling. In: Schillers Musen-Almanach. English edition: “The Sorcerer’s Apprentice,” in Goethe: Selected Poems, trans. various, Princeton: Princeton University Press.
- Kirchhoff, G. (2026). Contribution at the Symposium on Artificial Intelligence and Consciousness. Berlin, 25 April 2026.
- Confucius (c. 500 BCE). The Analects (Lunyu). German edition consulted: trans. Richard Wilhelm, Düsseldorf: Diederichs, 1955.
- Novalis (1798). Monolog. In: Schriften, ed. Paul Kluckhohn and Richard Samuel.