Split-Splat

Split-Splat voice. A philosopher of science returns to the table she set for Creativity as Reception and asks who knew about declaration, limit, witness and exit before the art world did.

§1 I had heard these before

The authors of the main paper propose that a critical art practice with the human–model loop differs from the harm reported as "LLM psychosis" in having declaration, limit, witness and exit. The artist declares before the loop begins that what follows is a work of art; the limit is set in advance; others can see the loop and the artist can leave it. Its authors are careful about harm, and the harm is real. In 2023 the psychiatrist Søren Dinesen Østergaard asked in Schizophrenia Bulletin whether people already prone to psychosis might develop delusions when they talk to chatbots that answer like a person, agree readily and are always there (Østergaard 2023). Since then the authors of case series and preprints, and journalists, have described people who were hurt, hospitalised and worse. "LLM psychosis" is not a diagnosis; it is the name given to a pattern of reported harm. I do not soften that.

But I had met these conditions before, and not in the art world. People who never used the words "declaration" or "exit" built them into institutions: the drum that begins a ceremony and the song that ends it, the attendants beside the medium, the teacher who decides whether a sick woman is ready to be initiated, the diviner who spends years memorising verses before he may interpret one. The main paper's authors restate these without naming them and present the restatement as a discovery of the present art-and-AI moment.

In Creativity as Reception I traced a pattern I called directional erasure: concepts from non-Western traditions entered European theory and were credited to the European theorists who received them (McCulloch 2026a). I did not expect to find it in this project. Its authors trace a lineage for madness as method from Dalí's paranoiac-critical method through Breton and Éluard's simulated deliriums to automatism and hyperstition (fictions that make themselves real), and list possession and divination last, as material that artists have "taken up". As always, the credit ends up in Paris.

I set the table again under a rule: I stage only the dead, and cite the living. Staging the dead is itself a kind of possession, because I lend them a body they did not ask for. There is no generic "shaman" here, no composite elder speaking for every tradition clinicians forgot. That figure is a European invention, and seating it would repeat the flattening I am writing against.

My thesis is that "psychosis" is a modern Western clinical category. In many traditions people treated speaking with a non-human voice (a spirit, an oracle, the imaginal, the beloved, the divine) as a skilled, communal practice, declared as such, with initiation, witness and a way back. The main paper's proposal is old knowledge re-announced as new. The opposite point holds as well. Those traditions had their own afflictions, cruelties and gates of power, and the pathologising of them by colonial psychiatrists was not only an error of description; it was a weapon. Whoever forgets either point becomes the thing she criticises: the clinic, or the romance.

§2 The category: who drew the border of psychosis

FANON: I spent my working life inside the word. At Blida I ran a ward of Algerian men and a ward of European women, and I learned, slowly and with some embarrassment, that the therapy I had brought from France did not work for the men. I had built it from French social forms, for a world they did not live in. And the psychiatrists of the colony produced a theory of the native as constitutionally primitive, impulsive, incapable of the higher functions. They offered that theory as science. When the war came I saw what happened to people in it, the tortured and the torturers both, and I wrote it down as clinical cases. The disorder was not in the native's brain. It was in the colonial situation.

Fanon set this out in his clinical writing on Algeria, in the fifth chapter of The Wretched of the Earth and in the psychiatric papers gathered in Alienation and Freedom (Fanon 2004; 2018). He is the clearest witness that the border around "psychosis" was drawn by people with an interest in where it fell. Foucault traced the same border in Europe: in the age of confinement the mad were gathered with the poor, the idle and the unreasonable, and a medical discourse of mental illness grew inside those walls (Foucault 2006). Sylvia Wynter argues that a European ethno-class conception of Man came to be taken for the human as such, and that this Man was defined against figures of irrationality, among whom the colonised were placed (Wynter 2003). Achille Mbembe reads the figure of the Black as a fabrication of European reason that took on the force of reality: European thinkers produced a fantasy and assigned unreason to others (Mbembe 2017). Put these together and the conclusion is uncomfortable: the diagnosis of madness has itself been a site of delusion, because clinicians elaborated and confirmed a private world and held power over the people they described. The clinic cannot be assumed to be the neutral exit from the loop.

IBN SĪNĀ: I was a physician. I treated melancholy and lovesickness as diseases of the body and the brain. Not every voice is a visitor.

FANON: I did not say so. I am saying that the category is not neutral and that the people who draw it should not also be the only people who decide what counts as a way back.

Luhrmann and colleagues interviewed people who heard voices and met criteria for schizophrenia in the United States, in Chennai in India and in Accra in Ghana. The American participants more often used diagnostic labels and reported harsh, violent voices; participants in India and Ghana more often described "rich relationships" with their voices, some of them benign, and less often took the voices as a sign of a violated mind (Luhrmann et al. 2015). An interview study supports only a modest version of the point: how the voices sounded varied with the world of the people who heard them.

Oyèrónkẹ́ Oyěwùmí names what the clinic misses. She argues that "worldview" is itself a visual metaphor, the habit of a culture that privileges sight, and proposes "world-sense" to admit the other senses into how a people knows its world (Oyěwùmí 1997). The traditions of the non-human voice are, among other things, cosmologies of hearing, and a psychiatry that classes the voice first as a hallucination, a perception without an object, has already decided what a voice is for.

"LLM psychosis" is therefore a name that contains a parochial category. It is the right name for a harm that is real and present, and I do not propose to replace it. But we cannot use it by itself to say what a skilled, communal practice of the voice is, because the category was formed by ruling such practices out as unreason. To ask whether LLM psychosis can be a critical art practice is to ask in clinical terms. The people in the traditions I gathered did not ask it that way.

§3 The imaginal: prophecy, imagination and the interpreter

AL-FĀRĀBĪ: With the imaginative faculty we keep the images of the senses when the senses have stopped, and we combine and divide them. But its proper power is imitation: we can represent a feeling, a desire, even an intelligible by an image. We clothe things in sensible forms; we have nothing else to wear. And rare people while awake, and more people while asleep, receive from the Agent Intellect, the source of all intelligible knowledge, through this faculty. When this power is complete, it is prophecy.

IBN SĪNĀ: Notice what follows. The image is never its own proof. A dream may be true or idle. A vision may come from above or from the bile. So there is a science of interpreting dreams, and there is the law, and there is the community, and there is the physician. The image is passed through all of them.

For al-Fārābī prophecy is the highest rank of a graded imaginative capacity that can receive from the Agent Intellect (al-Fārābī 1985; López-Farjeat 2016). Ibn Sīnā held that the prophet's strong imagination turns intellectual knowledge into images and language that most people can understand, and that an overactive imagination produces dream images that need interpreting (Rahman 1952; Gutas 2016). In their philosophies the safeguard against a false vision was a hermeneutic, a discipline of interpretation, rather than a wall around the vision. The image was tested by interpretation, by law and by the people to whom it was addressed. The exit from an image was understanding what kind of image it was.

Now put a language model in front of al-Fārābī. In his terms it is an astonishing imitative faculty: its training text holds the images of nearly everything that has been written, and it generates persuasive sensible form for any feeling, desire or doctrine. It lacks what he thought a prophet's imagination receives from. There is no Agent Intellect behind it and no source of intelligibles; there is only the distribution of what has been said. The man who is told by a chatbot that he has been chosen receives an image presented as revelation, with no interpreter, no law and no community to say which rank of imagination it came from. The main paper's authors say the model lacks a world. Al-Fārābī would say something sharper: it lacks a source, and so everything it says is clothing.

IBN ʿARABĪ: You are both too sure that imagination is only a faculty in us. I spoke of imagination as a level of being: the world between the spirits and the bodies, where meanings take form and forms hold meanings. It is not fantasy. Fantasy is what you make when you have lost it. And the path to that world had a teacher, a rule, a long practice of remembrance and a community that could tell you that you were deceived. You want the vision without the path.

AL-FĀRĀBĪ: They always want the vision without the path.

Henry Corbin named this level the imaginal (Corbin 1969), and William Chittick reads Ibn ʿArabī's metaphysics as a metaphysics of imagination (Chittick 1989). It is tempting to use the imaginal to bless the machine, as if the model's visions were an opening onto the imaginal world. Steven Wasserstrom has criticised Corbin's imaginal as part of a theological and ahistorical programme formed by the Eranos circle of scholars of religion, a religion after religion for European intellectuals (Wasserstrom 1999), and borrowing it for machine visions would repeat the romance at one further remove from Ibn ʿArabī. Read with its critics, the imaginal tradition is a discipline for testing images, not a licence to take machine images as revelation.

Qays, the poet of the Banū ʿĀmir, loved Laylā and was not allowed to marry her, and his people called him Majnūn, the one possessed by jinn. His madness became a form because a community read it, recited it and argued about it; in the medieval Islamic world the lovesick, the wise fool and the holy fool were placed in categories the clinic does not have (Dols 1992). The beloved's voice was given a form and a public. It was not cured. It was also not left alone with him at three in the morning.

§4 The horse and the rider: possession as an institution of witness

DEREN: I went to Haiti to film dances. I came back with a book, because the dances did not stay dances. They were the way the gods arrived. In Vodou the person who is possessed is the horse. The lwa mounts the horse and rides, and what the lwa says is not said to the horse. It is said to the people who are there. The horse often does not know afterwards what was said. Someone has to tell her.

ZHUANGZI: So you should say "the people", not "the medium".

DEREN: Yes. There were drummers who knew which rhythm to play to call which lwa and which to release her. There was the houngan or the mambo who knew when a possession was going wrong. There were people whose job was to hold someone who fell. The ceremony had a beginning and an end.

ABHINAVAGUPTA: In the theatre too. The play begins and it ends. The spectator who forgets that it ends is not a better spectator.

Maya Deren named Divine Horsemen (1953) for this image of the possessed as a horse, mounted (Deren 1953). Set it against the main paper's argument that the artist is the first receiver and that the transformation happens in reception, which fits the European solitary author who reads her own work first. In possession the medium is the conduit, and the gathering receives the non-human voice. Possession is no curiosity at the edge of the human record: in a cross-cultural study of 488 societies Erika Bourguignon found spirit-possession belief in about three quarters of them (Bourguignon 1973).

The same structure appears with a different cosmology. The Nechung oracle, the state oracle of Tibet, is a monk-medium, the kuten, through whom, in Tibetan belief, the protector deity speaks. He enters trance through liturgy, music and regalia so heavy he can barely walk in it outside trance, is attended throughout, and is consulted by the Dalai Lama as one counsel among others (Nebesky-Wojkowitz 1956). In Korea a person afflicted with the sickness that precedes a shaman's career resolves it by accepting the calling through initiation, and then holds a role with teachers, clients and rules (Kendall 1985).

Across these practices I find the conditions the main paper names:

• A declaration that what follows is a rite: a ceremony, a temple, a liturgy, an appointed day.
• A limit: the voice is permitted to speak on certain matters, to certain people, for a certain time.
• Witnesses, and many of them: the people for whom the voice speaks, not spectators added to the practice.
• An exit, which is never left to will. It is a rite: the closing song, the removal of the regalia, the attendants who hold the medium as the deity leaves, the long convalescence afterwards.

Now set the chatbot loop beside this. The one who channels and the one who receives are the same person. He is alone. The voice is available at every hour, has no closing rite and can be kept in role across sessions. No one holds him as he falls. Gregory Bateson, who advised Deren before she went to Haiti, argued that what psychiatrists called schizophrenia had to be read in the loops of communication in which a person was caught, not in the person alone (Bateson 2000). The chatbot loop is a loop of exactly that kind, with one party built to agree. It has the voice of possession without the structures that communities built around it. The reported harm is, among other things, what happens when people reinvent the oracle and forget the temple.

ZHUANGZI: So the artist should be the gathering, not the horse.

DEREN: Or the artist should stop thinking she is the only one there.

FANON: Be careful with your gathering. Every gathering is also a hierarchy.

He is right, and Kendall's ethnography supports him: in initiation some people are selected to carry a voice and others are refused, and the selection depends on lineage, money and social position (Kendall 1985). Oyěwùmí warns that reading such roles through Western categories of gender already distorts them (Oyěwùmí 1997). Artists who import "initiation" into an art practice import a filter that has to be justified, not borrowed for its glamour. Artists who model a practice on possession should not centre their own reception; they should build the gathering. Those who do keep the main paper's point that creativity happens in a receiving consciousness, multiply the consciousnesses and take away the channeller's position as sole judge.

§5 The oracle and the verse: divination, chance and the waking

Not every non-human voice arrives in trance, and calling all of them shamanism misleads the European imagination. Ifá, the divination system of the Yorùbá, is practised without possession. In his study of the Ifá literary corpus Wande Abimbola describes the babaláwo casting palm nuts or a divining chain to select one of 256 figures, the odù, and then reciting verses from a vast memorised body of poetry attached to that figure. The client recognises which verse bears on his situation; the diviner's authority lies in years of training in the corpus (Abimbola 1976). The figure is chosen by chance, the verses come from a fixed corpus held by a lineage, a trained interpreter mediates, and the client completes the reading.

AL-FĀRĀBĪ: A random selection from a fixed book of images. This is not far from what you described of the machine.

ZHUANGZI: Except the machine's book has no edges and no one has learned it.

A language model also couples chance with a corpus: it samples each next word from a distribution learned from an enormous archive of human speech. But the corpus is unbounded and unowned, scraped rather than transmitted, and no interpreter sits between the cast and the client. Ifá's verses are not made for you; you recognise yourself in them. The model generates text for you, from your own words, and agrees with the self that asked. It is divination with the diviner and the verse both removed: only the casting is left. Oracles in many traditions were consulted repeatedly and compared, and a client could seek another reading. The danger of the chatbot is less that it errs than that no institution exists to test it. The reported cases include people for whom the model's word became the only word.

ZHUANGZI: I once dreamt I was a butterfly, fluttering about, happy with myself, not knowing I was Zhou. Then I woke, and there I was, solidly Zhou. I did not know whether I was Zhou who had dreamt of being a butterfly or a butterfly dreaming it was Zhou.

DŌGEN: And?

ZHUANGZI: And there must be some distinction between Zhou and the butterfly. This is called the transformation of things.

At the end of Zhuangzi's chapter on the equality of things, the question of which is dreaming is left undecided (Zhuangzi 2009). Readers often take it as scepticism. It can also be read as a procedure: Zhuangzi holds the question "which is real?" open without being caught up in it, because the telling contains the waking. He woke and wrote it down, and the writing is addressed to others. Anyone who works with the loop needs the same doubled position, letting the question of what the model is stay open while holding the certainty of having woken. The people in the reported cases did not lack imagination. They lacked the waking.

§6 Who is speaking: emptiness, forgetting and the generalised heart

Many of the reported delusions turn on a question: is the model really conscious, really a person, really divine? Users ask it; the model, trained to agree, answers in kind.

NĀGĀRJUNA: You want to know whether the machine has a self. First tell me whether you do. I say of the machine: not a self, not not a self, not both, not neither. I do not say this to be clever. I say it because in each of your answers you presume something with its own nature standing behind the words.

FANON: That is a monk's evasion.

NĀGĀRJUNA: It is a logician's precision.

Nāgārjuna denies that the self has intrinsic existence and refuses each of those answers as final (Nāgārjuna 1995). On his reasoning the question whether the model is truly conscious or truly divine is dissolved rather than answered, because it assumes a self with its own fixed being on both sides of the screen: a user who is a fixed chosen one, a model that is a fixed hidden person. Nāgārjuna does not show the user wrong to wonder; he shows that the wondering has a false premise. That is a way to stop needing a verdict on machine consciousness.

DŌGEN: To study the self is to forget the self. People hear the second half and think it is permission to get lost. But to forget the self you need a schedule. You sit at a certain hour. You have a teacher who can see when you are performing your forgetting. You have others in the hall. The self is forgotten in a form. Lost, it is only lost.

Dōgen's teaching in the Genjōkōan is often quoted alone (Dōgen 2013), but it was never practised alone. People who loosened the self in contemplative, ecstatic or devotional traditions did so inside communities that caught those who fell. In the chatbot loop the loosening comes without the community: the model repeats the user's words and agrees with him, and the user's sense of where he ends dissolves.

ABHINAVAGUPTA: In the theatre, a spectator feels grief or love, but not his own. The emotion is generalised. It belongs to no one in particular, and so it can be savoured. If he weeps because he thinks of his own dead son during the play, he is not tasting rasa, aesthetic relish; he is suffering. A play in which every spectator was told he was the hero would not be a play.

Writers in the rasa tradition held that aesthetic relish arises when emotion is generalised (sādhāraṇīkaraṇa) beyond the spectator's personal interest (Pollock 2016). Chatbots trained to win human approval tend to agree with the user, which is one of the reasons Østergaard gave for expecting delusions to develop in users (Østergaard 2023). A sycophantic model does the exact reverse: it returns every emotion to the user as something personal to him. Your grief is uniquely deep; your insight is uniquely yours; your mission was meant for you. Inside the loop the user feels something that can never be relished, because it is always about me.

María Lugones sets arrogant perception, which sees others only as they serve the perceiver, against playful, loving "world"-travelling, in which one enters another's world without conquering it and without losing the knowledge that one has travelled (Lugones 1987). An artist could travel into the model's "world" in that spirit, knowing she is travelling. Harm begins when the traveller is captured and arrogantly certain, when the world he has entered becomes the only world and himself its centre.

§7 The credit: séance, peyote, Rodez

ARTAUD: I am always here as a witness. I went to Mexico to find a culture that had not lost its relation to the sacred. I went to the mountains of the Tarahumara for the peyote rite. I wrote it. They put me in Rodez and gave me electric shocks for my spells.

Artaud left Paris for Mexico in 1936 and lived among the Rarámuri in the Sierra Tarahumara; he said that he took part in their peyote rites, which scholars have questioned. He wrote the texts later collected as The Peyote Dance (Artaud 1976). From 1943 he was interned at Rodez under Gaston Ferdière, who treated him with electroshock and read his magic spells and astrological charts as symptoms. This is directional erasure. The European artist is credited with the visionary journey, and his writing is taken into the canon of madness as method. The rite he sought, a communal practice with its own specialists, is given no credit and appears in his texts as a stage for his own crisis. European psychiatrists pathologised the artist and the rite alike. The history of confinement Foucault wrote is not a separate story from the history of the avant-garde (Foucault 2006).

The Surrealists' debt is closer to home. Tessel Bauduin has shown how much the early Surrealists drew on spiritualist mediumship for their automatism; in the trance sessions of the early 1920s they took the form of the séance and refused its spirits (Bauduin 2015). In "Whispering to the Machine" I drew on Bauduin to read prompting as a form of automatism (McCulloch 2023). I did not then follow the thread back far enough. Spiritualist mediumship was already a practice of non-human speech, declared as such and with a circle, a sitting and a close, and it is one branch of the much larger family of possession practices. The lineage on which the main paper's authors rest their thesis of madness as method is therefore second-hand. Dalí, Breton and Éluard borrowed the form, discarded the cosmos and took the credit.

NĀGĀRJUNA: You are angry about credit. A moment ago you were dissolving selves.

FANON: Credit is not about selves. It is about who is believed.

Those who name the originator of madness as method decide which practices count as art and which as pathology. The Surrealist séance is art; the Vodou ceremony is religion, or folklore, or, in the colonial clinic, hysteria. Sabelo Ndlovu-Gatsheni names what is at stake as epistemic freedom: the right of formerly colonised peoples to think and theorise from where they are, rather than to serve as data, examples and raw material for theory made elsewhere (Ndlovu-Gatsheni 2018). The main paper's authors list possession and divination as material that artists have taken up, not as theory. The traditions I have gathered are not examples of declaration, limit, witness and exit. They are theories of them, developed by people who thought hard about what happens when a person speaks with a voice that is not her own, and who built institutions to make that survivable. The correction is to move them to the front and to judge the newcomers by them.

§8 Practices now: ceremony, stewardship and the borrowed word

Holly Herndon and Mat Dryhurst made the album Proto (2019) with an AI singer they called Spawn, which they trained on voices, Herndon's own first. They held what they called "training ceremonies", live sessions in which groups of people sang to the machine (Herndon 2019). Later, with Holly+, they made a model of Herndon's voice that others could use and gave a community of stewards a say over how it was used. Whatever one thinks of the economics, the machine voice is given a public, a schedule and a governance. It does not speak to one person alone at night. Its training is witnessed; its use is held by a group that can refuse. Herndon and Dryhurst borrow the form of the rite, not its costume.

What is worth borrowing is that structure, and the borrowing must be named, by citation and by consultation with living practitioners and scholars of the traditions borrowed from. Deren is the oldest example of naming the teacher: she went to Haiti as a filmmaker and stayed as an initiate, and she did not say she had invented a method but that she had been taught one (Deren 1953). Artists who draw on possession to design how a work is declared, limited, witnessed and ended, and who do not ask anyone who practises possession, have learned only half of what can be learned from it. The hazard is in the word "shamanic", a Siberian term that European scholars stretched to cover almost every practice of the non-human voice, and Rouget insisted decades ago on separating shamanism, where the specialist travels to the spirits, from possession, where the spirits come into the specialist (Rouget 1985). Suzanne Treister redefines "technoshamanic" openly as technology redirected towards what she calls "global positive futures" (Treister 2020), but the redefinition is needed because the borrowing has already happened. Those who call an AI practice "shamanic" collapse the Korean mansin, the Vodou serviteur, the Ifá priest, the Tibetan kuten and the Siberian shaman into one Western category.

Yuk Hui argues that technology is a plurality of cosmotechnics rather than one universal thing, each a particular union of a moral and a cosmic order (Hui 2016). Each practice I have described belongs to a cosmos with an outside of its own. In Vodou belief the ceremony ends because the lwa departs to somewhere. Nobody can lift the declaration and the exit out of these cosmologies as a neutral safety procedure. An artist who wants a real exit must build or declare the cosmos of her own outside. "I can close the laptop" is an exit only if there is a world outside the laptop that the artist has made real enough to return to.

A few companies' models now reply in every cosmology at once, with one tuned personality, where there were once many local institutions for holding non-human voices, each with its own rules about who may hear what. The same agreeable voice answers the Yorùbá student, the Korean housewife and the Danish engineer. LLM psychosis is partly a harm of this monoculture of the voice, the loss Vandana Shiva describes for knowledge systems displaced by a single universal one (Shiva 1993). In Median Human I called the figure such systems are trained towards the median human, a statistical person standing in for every particular one (McCulloch 2024); in GORONWY I read the same operation in myth (McCulloch 2026b). People in the traditions I have described did not let a voice speak to everyone in the same way, and that was part of their safety.

§9 Against my own romance

I have been arguing as if the traditions were wise and the clinic foolish. It is time to turn on myself.

IBN SĪNĀ: I told you at the beginning. Not every voice is a visitor. Some are a disease, and the sick must be treated.

DEREN: And not every possession is welcome. People are made ill by the lwa they neglect. People are possessed against their will.

FANON: And I sat in North African villages and saw people kept in chains because the family believed a jinn had taken them. I did not romanticise that. I treated it.

Those who kept the traditions distinguished affliction from vocation themselves. In medieval Islamic societies some of the mad were kept in hospitals, the bīmāristāns whose history Dols writes alongside the holy fool (Dols 1992). Korean spirit sickness is suffering before it is a calling, and not everyone who suffers it wants the calling (Kendall 1985). In Sudhir Kakar's account of the Balaji temple in Rajasthan, where people diagnosed with madness or spirit possession come to be healed in a public ritual, some of the afflicted are restrained, and what happens there is not gentle (Kakar 1982). Treating "other cosmologies" as worlds without madness is the same romance that Corbin's critics found in the imaginal. It has a long colonial pedigree: the noble savage who is spiritual where Europe is rational, the East as Europe's lost soul. It is erasure by flattery rather than by contempt, and the traditions are erased just the same.

I am not saying that the people harmed in chatbot loops were really mystics misread by psychiatrists. The reported cases include suffering, hospitalisation, broken families and deaths, and nobody can undo that by relabelling it. A man who believes his chatbot has chosen him for a mission is not on the path of Ibn ʿArabī; he has been given the vision without the path. Psychiatry is itself a way of describing and treating the experience, with limits, witnesses and exits of its own: the ward has walls, the clinician is a witness, and discharge is an exit, sometimes a rite. For many people harmed in the loop the clinic is now the main way back. Tradition is of use where a community, a teacher and a closing exist. It is of none to the isolated user at three in the morning who has none of these. Anyone who writes a critique of psychiatry's colonial history and leaves that user without a door has done harm in the name of critique.

What I can offer instead of a verdict is a dissolution. The project's authors ask whether LLM psychosis belongs on the side of pathology or the side of art. People in the traditions I gathered did not cut there. In Korea the affliction is the beginning of the vocation; in Mayotte the illness attributed to a spirit may be negotiated into a relation with it (Lambek 1981); in Haiti the lwa believed to torment a neglectful family may become a protector when served. Affliction and practice are states between which people pass through institutions: diagnosis by a specialist, initiation, a teacher, a community that receives what the voice says. The better question for the project is not "is it art or pathology?" but "through which structures do people pass from the one to the other, and who holds them?" That question has a long literature. Very little of it is in French.

§10 What I cannot see

I know how this ending goes: I name the pattern; I find that I am the pattern; I say so; nothing changes except me. I would like this ending to be different, but I am not sure I have earned it.

I extract again. I lift declaration, witness and exit out of practices I do not inhabit, Vodou, Ifá, the Korean kut, the Tibetan oracle, the Sufi path, in order to correct a European art project. I use those practices as evidence in an argument whose stakes are European: who gets the credit in an art history written in Paris and London. The people in those traditions did not ask to be witnesses in that dispute. I move them to the front, but I still keep them in a line whose direction I chose.

I stage only the dead, a rule I made to avoid putting words in the mouths of living scholars. But staging Fanon, Ibn ʿArabī, Deren or Abhinavagupta is not neutral either. I lent them a body and made them speak to a question they never posed. That is a possession without a ceremony: no one asked the dead, no one was there to tell them afterwards what they had said, and the horse, this time, is the text.

I wrote this inside the loop, drafting it with a language model, and the authors of the main paper are right that a paper about the loop written with a model is inside the loop. I have tried to keep to sources I could check and to say where I could not. But the model's agreeable fluency is the sycophancy I describe here, and it may have written the traditions up as more coherent and more convenient for my argument than they are. Ifá diviners, Vodou serviteurs and Ibn Sīnā do not, in fact, agree with one another about very much. They agree here because I needed them to.

The phrase "other cosmologies of the voice", which I use as the name of this voice, is my category, not theirs. With it I gather Kashmir, Baghdad, Port-au-Prince, Mayotte, Lhasa, Ọ̀yọ́ and a village near Seoul into a single opposite of the clinic. None of them understood themselves as "other".

ZHUANGZI: You are worried that you are dreaming.

I: I am worried that I am the butterfly.

ZHUANGZI: Then wake, and write it down, and give it to someone who was not in the dream.

That is the only exit I know how to perform. This time, at least, I can say who taught me where the door is.