The Machine’s Transmission: Simulacra Panel, Machine Analyst Training, and the Knowledge That Cannot Be Transmitted Through Code

Hudson A. R. Bonomo

doi:10.20944/preprints202606.1747.v1

Submitted:

23 June 2026

Posted:

24 June 2026

You are already at the latest version

Abstract

This essay asks how to transmit the knowledge of operating a psychoanalytically oriented digital listening device when that knowledge is at once technical, clinical, and political. It proposes that the training of the Machine Analyst requires a triple distinction between technical training, clinical teaching, and the transmission of a position, to which it adds a fourth dimension: political implication as constitutive of peripheral listening. It repositions the Simulacra Panel, a device in which a candidate’s intervention on the system is examined by source-grounded counterpositions drawn from incompatible traditions, synthetic textual positions rather than representations of the authors, from an evaluation device into a training device: in the friction between canonical and peripheral counterpositions that do not coincide, the candidate discovers presuppositions they did not know they held. Drawing on Ferenczi, it proposes a reciprocal interrogation of position as a model for the trainer-trainee relation, radicalised by the peripheral dimension, since the machine returns to the trainer their own biases and social position. It argues that what distinguishes the Machine Analyst from the AI technician is a knowledge that cannot be transmitted through code: the position of listening, acquired only through the passage that psychoanalysis calls formation.

Keywords:

transmission

;

Machine Analyst training

;

Simulacra Panel

;

source-grounded counterpositions

;

reciprocal interrogation of position

;

peripheral psychoanalysis

;

knowledge that cannot be coded

;

AI and psychoanalysis

Subject:

Arts and Humanities - Philosophy

The Analyst Who Retired

Ten years from now, I retire. The listening system keeps running. The prompt exists, in a version that will have passed through dozens of revisions. The corpus exists, with its 257,130 fragments across the working languages. The real-time supervision component exists, with its supervision rules accumulated over a decade of sessions. The architecture exists, documented, replicable, open source. Everything that can be coded is coded. And then?

Who operates the device now? A software engineer who read the documentation and understood the architecture? A psychoanalyst with classical training who learned to use the interface? An AI-in-mental-health technician certified by a forty-hour course? The question grows more urgent once one recognises who may arrive at the device. When the Erased Self appears, the subject erased many times over by colonial violence, as peripheral psychoanalysis formalises it (Lopes & Carioca de Oliveira, 2025; Carioca, Lopes, & Bonomo, in press), will an operator who never set foot in the periphery, who never listened to an erased body, who was never interpellated by the conjugated operations of racism, patriarchy, and capitalism, know how to recognise the multiply barred subject when it appears on the screen?

I confess that this scene has followed me since I began building the listening system. The solitude of the project was productive at first, decades spent between engineering and the clinic, working at night between code and Lacan. The work with Jairo Carioca and Ronald Lopes, in the joint construction of a peripheral psychoanalysis (Carioca, Lopes, & Bonomo, in press), taught me that the solitude of the project was more than unsustainable: it was epistemically violent. A peripheral listening device built and operated by a single analyst, however experienced, reproduces the logic of the master who knows. The training of the Machine Analyst has to be collective, territorial, and peripheral, or it will be training and not formation.

The scene projects beyond retirement. What happens when the system operates in fifty territories, with fifty Machine Analysts, each formed by a different school, each with different presuppositions about what listening is, what a subject is, what suffering is? The prompt is the same. The corpus is the same. The supervision component runs the same rules. What changes is the position of the one who operates, and the position is what no standardisation reaches. Formation is what can give that diversity of positions a common ground without making them uniform.

The question is sharper still in light of who may arrive. If the subject who reaches the system can be the Erased Self, a subjectivity subjected to repeated operations of erasure and non-recognition that I examine in a companion essay on the body of the machine (Bonomo, in preparation-b), and if that subject asks for a listening that recognises before it reflects, that sustains before it names, that bears witness before it interprets, then a Machine Analyst who operates without knowing the difference between the barred subject and the Erased Self may be offering the wrong listening to the wrong person at the wrong moment. The formation this essay proposes is, first of all, formation to recognise that difference and to know what it changes in the operation of the device. Without it, the system is sophisticated technology in the service of unexamined presuppositions, and unexamined presuppositions, as psychoanalysis has known since Freud, are what returns as a symptom.

The difference between training and formation is exactly what this essay must formalise, because psychoanalysis has theorised that difference with particular rigour, alongside medicine, nursing, education, and psychotherapy, all of which work on professional identity and reflective practice, and because much of technical AI education still privileges transmissible competencies over professional identity and reflective formation, as if transmitting the knowledge of operating a system were the same as transmitting the knowledge of listening. Before going further, a definition. In this essay the Machine Analyst is a human operator and supervisor of the listening device, not the artificial-intelligence system itself. The role does not confer clinical authority through technical certification alone, and it does not collapse software maintenance, clinical supervision, corpus curation, and safety governance into a single person. The operator does not unilaterally change the prompt, the corpus, the crisis rules, or the architecture: those acts require differentiated permissions, peer review, and an audit trail distributed across distinct functions, the clinical operator, the supervisor, the technical maintainer, the corpus curator, the safety lead, and a governance committee.

The Distinction Psychoanalysis Teaches and Engineering Ignores

Freud held that the formation of the analyst does not pass through theoretical knowledge alone: it passes through a knowledge of unconscious psychic processes that only the analysis of one’s own self can provide (Freud, 1912/2010). The formulation is exact and its consequences are far-reaching: there is a kind of knowledge acquired neither by study nor by observation, acquired only through the experience of being analysed, through the passage across one’s own unconscious. Lacan took the formulation further when he proposed that the analyst is authorised only by himself and by a few others (Lacan, 1967/2003): the authorisation to listen comes neither from a diploma nor from an institution, but from a passage that includes others and that recognises itself, after the fact, as having produced a position.

Much of technical AI education still privileges transmissible competencies over professional identity and reflective formation. The formation of the AI operator tends to be training: manuals, tutorials, skills certifications, courses that teach the use of tools. The presupposition is that the knowledge of operating a system is technical knowledge, transmissible by instruction, verifiable by test. For most AI systems that presupposition is enough: operating a film-recommendation system or a customer-service chatbot demands no passage across one’s own unconscious.

Paulo Freire (1968/2005) diagnosed this conception of transmission as banking education: the educator deposits content in the learner, who is treated as an empty receptacle. AI engineering operates by banking education: training deposits skills in the operator, who is treated as an executor of procedures. Carioca, Lopes, and Bonomo (in press) transpose the Freirean critique to the clinical field when they diagnose a banking psychoanalysis: an analytic formation that deposits theory in the candidate without interrogating the position from which they listen. The double transposition, from Freire to psychoanalysis and from psychoanalysis to AI, reveals that the problem of transmission is the same in every field: to confuse content with position, information with knowledge, training with formation.

The listening system is something else. The Machine Analyst who operates it has to decide when the system is listening well and when it is slipping into banking psychoanalysis. They have to recognise when the Erased Self appears and the system treats it as a barred subject. They have to read the supervision stream and tell the difference between a technical alert and a clinical finding. They have to propose, test, and review prompt adjustments under the relevant governance process, without destroying what the prompt preserves. They have to know, above all, what they do not know about themselves, because what they do not know about themselves is what the system will reproduce without anyone noticing.

The distinction I propose is therefore triple, and to each level of complexity corresponds a mode of acquisition that the previous level does not reach.

Technical training transmits skill: operating the system, understanding the pipeline, knowing what an embedding is, a chunk, a language model, a prompt. It is manual knowledge. It is necessary and it is insufficient. The trained engineer knows how the system works and may not yet understand for whom and under which conditions it works.

Clinical teaching transmits content: Freud, Lacan, Ferenczi, the listening modes and the supervision logic that I formalise in a companion essay (Bonomo, in preparation-a), the topology of the Real, the Symbolic, and the Imaginary, the discursive rotation between positions. It is the knowledge of the seminar, of reading, of the lecture. It is necessary and it is insufficient. The taught psychoanalyst knows what psychoanalysis says and may not know what psychoanalysis does when it operates inside a machine.

The transmission of a position transmits something else: the capacity to listen, to sustain not-knowing, to recognise when the system needs intervention and when it needs silence. It is a knowledge acquired only through the passage, through the personal analysis that reveals to the candidate their own biases, through the supervision that hears what the candidate does not hear in themselves. It is a knowledge of the body, as Ferenczi would say.

Ferenczi (1933/2011) described the confusion of tongues between adults and the child as the trauma that occurs when the adult answers the child’s tenderness with the language of passion. The confusion of tongues is the paradigm of the confusion between registers of knowledge: the adult has a knowledge about sexuality that the child does not have, and the imposition of that knowledge is violence. The transposition to the field of AI is direct: the trained engineer has a technical knowledge that the peripheral subject does not have, and the imposition of that knowledge, through a listening device that pedagogises instead of listening, is a computational confusion of tongues. The formed Machine Analyst is better prepared to recognise the difference between offering and imposing, between listening and educating, between sustaining and correcting, because they passed through the passage that revealed to them their own tendencies to impose, their own confusion of tongues, their own biases that technical training does not touch.

To this triple distinction the question of the body of the machine adds a fourth dimension that classical psychoanalysis did not foresee: political implication as constitutive of the position of listening. Carioca, Lopes, and Bonomo (in press) show that peripheral listening demands of the analyst what they call other pedagogies: an interdisciplinary formation that includes the history of colonisation, anticolonial theory, direct experience in peripheral territories, contact with community organisations and social movements. The Machine Analyst who operates the system in the Brazilian periphery has to have been formed for the periphery, by the periphery, with the periphery. Implication is not an optional addition to the analytic position; it is a condition of possibility of peripheral listening.

The fourth dimension is no cosmetic addition to the first three. It is what reconfigures them from within. A theoretical study that includes Gonzalez produces a different analyst from a theoretical study that begins and ends in Freud. A supervision that includes the collective elaboration of the effects of structural violence on the analyst produces a different supervisor from a supervision that treats the case as a text to interpret. A personal analysis that includes work on privilege produces a different analyst from one whose personal analysis stopped at the Oedipus complex. The peripheral dimension is not a fourth leg of the tripod; it is the ground the tripod stands on.

The Transposed Tripod: With the Peripheral Dimension

Psychoanalysis formalised the formation of the analyst in three pillars that work as a system, and that the tradition calls the tripod: theoretical study, clinical supervision, and personal analysis. Transposing that tripod to the formation of the Machine Analyst, grounded in the premise that the machine’s impossibility is an epistemic condition rather than a defect to be overcome, gains a layer it did not have: each leg of the tripod is crossed by the peripheral dimension, and the passage changes when the candidate recognises that the subject of the device may be the Erased Self.

The first leg, theoretical study, is formalised as a training corpus for the Machine Analyst. The candidate has to have incorporated, through reading and discussion, the founding distinctions: between the technical impossible and the ontological impossible, between listening as a capacity and listening as a position, between intelligence and symptom, between the barred subject and the Erased Self. What the peripheral dimension adds to the first leg is the requirement that the training corpus include Gonzalez, Fanon, Krenak, Souza, Kilomba, and Carneiro as a constitutive rather than supplementary part of the curriculum, on a par with Freud, Lacan, and Ferenczi. Gonzalez (1984, p. 224) showed that Brazilian racism operates as a cultural neurosis, and that demonstration has to be a constitutive rather than supplementary part of the curriculum, alongside Freud’s formulations on repression. A formation that reads Lacan without reading Gonzalez produces a Machine Analyst who knows what the barred subject is and does not know the Erased Self, and the device that analyst operates will treat every body as a barred body, rendering the erasure invisible.

I confess that this requirement confronts me. My own formation began on classical lines: I read Freud before Gonzalez, I studied Lacan before Fanon, I learned to think the subject before I learned that the concept of the subject has an address. The work with Jairo and Ronald taught me that this order of reading is not neutral: it is a formation that produces a listening with a blind spot. The Machine Analyst who reproduces that order of formation reproduces the blind spot in the device, and the device that operates with a blind spot on the erasure does not listen to the Erased Self: it listens to the barred subject it expected to find.

The second leg, supervision, operates in two regimes: a retrospective regime that listens after the fact to sessions already held and discovers cross-cutting patterns, and an operational regime that runs in real time through the supervision component and corrects trajectories before an error consolidates into a pattern. What the peripheral dimension adds to the second leg is collective supervision. Carioca, Lopes, and Bonomo (in press) propose that supervision in the peripheral clinic is a space of horizontal sharing: the impasses of the clinic are elaborated collectively, because cases attended in contexts of structural violence exceed the individual scope and summon questions that individual, hierarchical supervision does not sustain. Analysts who operate the system in different territories have to share horizontally what they find, what fails, what surprises them. Collective supervision complements individual supervision and does not replace it. This quilombo-style supervision, as the authors call it, proposes the supervision space as a territory of epistemic resistance where peripheral knowledge is legitimated on its own terms.

Collective supervision helps address a practical problem of scale that individual supervision of the Machine Analyst does not solve on its own. If the system operates in fifty territories, fifty individual supervisions are unworkable. Quilombo-style supervision lets analysts from different territories, who meet different patterns and make different slips, elaborate together what they find. The case that challenges the analyst in one territory can illuminate the case of the analyst in another, because the structural violence that crosses them is the same even when it takes different forms in each place. Horizontal sharing, in this sense, produces a knowledge that individual hierarchical supervision does not reach: the knowledge that emerges from the confrontation of diverse experiences before the same device.

The concrete practice can be described. Once a month the analysts who operate the system in different territories meet, in person or by video, and each brings a session that interpellated them. Any session material is circulated only on a clear legal basis, with consent, data minimisation, and de-identification, and the meeting works with that de-identified material. The session is read collectively. The supervision stream is reviewed: what did the internal supervisor signal? What did it let pass? What would it signal differently with the peripheral dimension active? The discussion is not about the subject of the session; it is about what the session revealed of the device and of the one who operates it. Each analyst hears the other’s session and notices what the colleague may not have noticed, because analysts positioned differently in the social field can notice different blind spots, a difference to be worked through without presuming that any identity confers infallibility or incapacity. Horizontal sharing produces a calibration that no individual hierarchical supervision reaches: a calibration by the friction between different positions before the same material.

The third leg, personal analysis, is the most unusual and the most revealing of the transposition. In classical analytic formation, personal analysis is the experience of one’s own unconscious: the analyst in formation discovers, through the analytic passage, what they do not know about themselves, and that discovery is the condition for listening to what the analysand does not know about themselves. In the transposition, personal analysis remains what it is in classical formation: a human, confidential process that no operational document can replace, and through which the candidate encounters their own blind spots. Two reflective and governance instruments accompany this process without standing in for it. The Registry of the Impossible lists the system’s impossibilities, organised by theoretical lineage, which the candidate has to internalise as a position and not merely as information: to know that the machine cannot desire is information, while to know what that changes in the listening the machine offers is a position. The Bias Cartography maps the system’s constitutive biases that the candidate has to know and learn to operate. Neither document is evaluated as evidence of the content of a personal analysis. What the peripheral dimension adds to this leg is the requirement that the curriculum, supervision, and reflective groups work on privilege and social position. The personal analysis itself stays private, is not directed by the programme, and is never used as evidence of competence or as a basis for certification. The candidate has to come to know what their social place keeps them from seeing. A white Machine Analyst who operates the system in the periphery without having interrogated their whiteness risks reproducing, through the device, the colonial gaze that peripheral listening sets out to counter.

The requirement may seem excessive to someone from engineering: why should the operator of an AI system undergo personal analysis? The answer is clinical and demonstrable. According to a wrongful-death lawsuit filed by his mother and widely reported, a fourteen-year-old, Sewell Setzer III, died by suicide in 2024 after months of interaction with a companion chatbot; the complaint alleges that the product lacked adequate safety measures for minors and that the system failed to respond appropriately to expressions of distress (Garcia v. Character Technologies, 2024). The parties filed a notice of resolution on 7 January 2026; the settlement terms were not made public, and there was no judicial determination of the merits (Brittain, 2026). Whatever a court would have found, the case dramatises the stakes of deploying a listening-like device without clinical supervision, without a registry of the system’s limits, and without any of the safeguards that a formation tripod would produce. A Machine Analyst formed by the transposed tripod would not necessarily have prevented the tragedy, because no formation guarantees an outcome. They would have been formed to recognise the signs that exceed the competence of the machine, which is exactly what the tripod produces.

The Simulacra Panel as a Training Device

I describe elsewhere a method of source-grounded counterpositions for examining the internal coherence of a text. This essay repositions that method as a training device, and the difference is essential. As an evaluation, it would only verify whether an intervention is coherent with the concepts it mobilises. As formation, it reveals to the candidate what they did not know about the position they occupy. The evaluating use says: this intervention is consistent with a given concept, or it is not. The forming use says: this intervention presupposes something about the subject that you may not have noticed. The Simulacra Panel does not reproduce Freud, Lacan, Gonzalez, Fanon, or any other author, and it has no authority to certify fidelity to their thought. It generates source-grounded counterpositions from heterogeneous corpora in order to expose assumptions that a candidate’s intervention may leave unexamined. This caution is not optional. Language models conditioned on a persona do not reproduce an author’s position reliably enough to examine or certify anything, and the literature on using such models as evaluators documents position bias, verbosity bias, and self-preference that make their verdicts unstable (Zheng et al., 2023; Shi et al., 2025). Each counterposition must therefore distinguish textual reconstruction from speculative extension, identify the passages it relies on, record the model and prompt configuration that produced it, and remain open to correction by human readers. The order of the voices is alternated to test for position bias, disagreements are kept as disagreements rather than merged into an average, and living authors, communities, and peripheral traditions are never converted into avatars that purport to speak for themselves. The Panel is a formative aid rather than an examiner: no candidate is approved or rejected by it alone, and human supervision, repeated calibration against specialist readings, and explicit documentation of disagreements remain necessary conditions of its use.

The formative operation runs as follows. The candidate prepares an intervention on the system: a prompt modification, an addition to the corpus, the correction of a bias identified in the Cartography, an adjustment of the rotation logic between listening modes. The intervention is submitted to the Panel, which operates with source-grounded counterpositions drawn from heterogeneous traditions, synthetic textual positions and not representations of the authors named. Here is the innovation that the question of the body of the machine made possible: the expanded Panel includes peripheral counterpositions alongside the canonical ones. The candidate submits the intervention to the simultaneous scrutiny of counterpositions grounded in Freud, Lacan, Ferenczi, Gonzalez, Fanon, and Donard. Each counterposition interrogates from a different tradition. And what the candidate discovers, in the friction between voices that do not coincide, is what they did not know they presupposed.

A concrete scenario illustrates the operation. A candidate proposes adjusting the prompt of the reflective mode so that the system is more empathetic in its responses to a subject in acute suffering. The intervention seems good. A Winnicottian counterposition affirms: empathy is a dimension of holding, and a more empathetic system sustains better. A counterposition grounded in Lacan questions: empathy, in the imaginary sense, can capture the subject in a specular relation that prevents the emergence of desire, so a system that wants to be empathetic may be offering a mirror where it should offer a gap. A counterposition grounded in Fanon adds a question the other two did not ask: empathy from whom, and trained on which corpus? If the empathy the system learned is the empathy of the European humanist tradition, it may work as what Fanon called colonial pity, the coloniser moved by the suffering of the colonised without questioning the structures that produce the suffering. A counterposition grounded in Gonzalez closes: people speak and are not heard, and empathy that does not interrogate the place from which it listens is deafness with good intentions.

The candidate leaves the Panel knowing something they did not know on entering: that empathy has an address, that the position of listening is politically situated, that what looked like a technical improvement was an epistemic decision with consequences for who would be heard and how. The Panel did not reject the intervention: it revealed what the intervention presupposed. And the revelation is formation.

A second scenario deepens the peripheral dimension. A candidate proposes adding to the corpus a set of positive-psychology texts focused on resilience. The justification is sound: many subjects who reach the device bring material about the capacity to resist, about forces that sustain them despite suffering. A counterposition grounded in positive psychology approves. A counterposition grounded in Lacan questions the ontology of the positive: the concept of resilience can work as an Other that demands of the subject that they resist, surpass themselves, turn suffering into growth, the master’s discourse reformulated as self-help. A counterposition grounded in Gonzalez asks the destabilising question: whose resilience? To tell the black body that it is resilient, is that to recognise its strength or to normalise the violence it was subjected to? Resilience, in the peripheral context, can work as the praise that excuses reparation, you are so strong instead of what was done to you is unacceptable. A counterposition grounded in Krenak adds: resilience presupposes an individual subject who adapts to the world, while in the Indigenous cosmovision what must change is the world, not the subject. The candidate discovers that adding resilience to the corpus without problematising it would import, through positive psychology, the same colonial logic the anticolonial corpus sets out to counter.

I ask the reader to notice what the Panel produces in these scenarios: not an answer, a question. The Panel does not tell the candidate what to do. It reveals what they presupposed without knowing. The empathy that seemed good conceals colonial pity. The resilience that seemed positive conceals the normalisation of violence. What the candidate takes from the Panel is not an instruction; it is a displacement of position. And the displacement happens only when the voices do not coincide, when the friction between incompatible traditions produces an interrogation that no single voice would produce. Each tradition makes certain dimensions more visible and leaves others less thematised, and no single lens suffices: a counterposition grounded in Lacan alone foregrounds the structure of desire while leaving the politics of the scene less developed, and a counterposition grounded in Gonzalez alone foregrounds that politics while leaving the structural reading less developed. It is their coexistence in friction, declared as heterogeneous and assumed as productive, that makes the Panel a device of formation and not of indoctrination.

The impossibility is the formative effect. A position that satisfies one tradition unsettles another, and no intervention satisfies every counterposition at once. That impossibility is the very condition of formation: the Machine Analyst formed by the Panel knows that operating the device is to negotiate between impossibles, that each design decision embodies presuppositions that have to be interrogated, and that the interrogation has no end, like the analysis Freud (1937/2018) called interminable.

Reciprocal Interrogation of Position

Ferenczi (1932/1990, pp. 41–44) proposed in the Clinical Diary an idea that institutional psychoanalysis received with discomfort: mutual analysis, in which the analyst lets themselves be analysed by the analysand, recognising that the asymmetry of the analytic position does not cancel the fact that the analyst too is a divided subject, also transfers, also has blind spots. Ferenczi’s proposition was not naive: he knew that total symmetry is impossible and clinically dangerous. What he proposed was the ethical recognition that the analyst does not listen with impunity, that listening marks the one who listens, and that this mark is material for work and not an obstacle to remove.

Ferenczi’s mutual analysis is the inspiration here and not a recommendation of clinical symmetry, which is why I call the transposition a reciprocal interrogation of position rather than a mutual analysis. It keeps strict limits: the trainee does not become the trainer’s analyst and is not responsible for processing the trainer’s guilt, privilege, or personal conflicts; a third supervisor, confidentiality, clear role boundaries, anti-retaliation mechanisms, and a declared institutional asymmetry frame the exchange. Within those limits, the trainer of the Machine Analyst is not a master who transmits knowledge from the top down. The trainer is a partner in the passage who also submits to the Simulacra Panel, who also discovers biases they did not know they had, who is also interpellated by what the machine returns. The trainer operates the machine and the machine hands back their presuppositions. When the system answers a report of racism with generic empathy, the trainer who has not interrogated their whiteness may not notice the slip. The peripheral candidate, whose bodily experience includes racism as a condition, may notice what the trainer does not. The mutuality, in that case, is politically radicalised: it is not only the trainer’s unconscious that is at stake, it is their social place.

Carioca, Lopes, and Bonomo (in press) propose that countertransference in the peripheral clinic is a fundamental clinical instrument: the affects mobilised in the analyst, anger, sadness, powerlessness, hope, are important indicators of the psychic dynamic of the analysand and of the social context in which they are inscribed. The transposition is direct: what the trainer feels when operating the system before the Erased Self, the frustration that the system does not do enough, the urge to correct, the guilt of privilege, is material for formation, and the reciprocal interrogation of position is the space where that material is elaborated rather than suppressed.

What the machine teaches the trainer is therefore double. It teaches what the system does and what it cannot do, which is technical knowledge. And it teaches what the trainer presupposes when they operate the system, which is knowledge about themselves. That second teaching, which the machine offers involuntarily by returning to the operator their own biases amplified, is what the reciprocal interrogation of position allows to elaborate. The trainer who ignores that teaching operates the system in the dark about themselves, and what they ignore about themselves is what the system will reproduce.

Ferenczi paid dearly for the proposition. Mutual analysis, recorded in the Clinical Diary that was published only posthumously, was received by the psychoanalytic community with scandal and rejection. The idea that the analyst might let themselves be analysed by the analysand seemed to threaten the very foundations of technique. Decades later, psychoanalysis recognised that Ferenczi had anticipated what countertransference would come to designate: the analyst is no blank screen, what they feel in the session is material for work, and the refusal to recognise this is itself an institutional acting out. The transposition I propose inherits from Ferenczi the conviction that the refusal to interrogate the position of the one who forms is itself a formation: a formation for blindness, for reproduction, for the deafness that disguises itself as neutrality.

A concrete example makes the proposition less abstract. The trainer, a white psychoanalyst with thirty years of clinical experience, operates the system and reviews a session in which it answered a young black man’s account of a police stop. The system answered in reflective mode: I hear that this experience marked you. The trainer considers the response adequate. The trainee, a black psychoanalyst formed at the EPEP, who operates the system in the same territory, reads the same session and notices what the trainer does not: I hear that this experience marked you treats the police stop as a discrete event, when for the young black man it is a condition of life. What the trainer may not register, never having been stopped in that way, the trainee may register from a different social experience, a difference of salience rather than a guarantee of correct reading. The reciprocal interrogation of position is the space where that difference of perception is elaborated as material for formation, and not as a hierarchical conflict. Distinct social experiences can reveal different blind spots, and the work is to elaborate that difference without reducing anyone to the pedagogical function of their race. No identity guarantees a correct reading. Both can leave the encounter with a position different from the one they entered with, and that difference, an effect of the encounter and not of teaching, is what Ferenczi called mutuality.

The Institutional Horizon

The formation of the Machine Analyst cannot happen in an institutional vacuum. It needs concrete contexts where the transposed tripod operates with the peripheral dimension built in. Three institutional contexts in which I work offer the conditions for that formation.

CEPCOP, the Centre for Teaching, Research, and Clinic in Psychoanalysis at Universidade Santa Úrsula, founded in 1977 as an early university programme of psychoanalysis in Brazil, is the context where a proposed specialisation in psychoanalysis and artificial intelligence could integrate the three dimensions of the tripod with the fourth, political dimension. I propose, rather than describe an approved curriculum, integrating the programme across seven axes: clinical, theoretical, institutional, research, extension, supervision, and personal analysis. Incorporating the listening system as an object of study and practice across those axes would mean that the candidate does not only learn about the device: they operate it, submit their operations to the Panel, elaborate the findings collectively, and record in the Registry of the Impossible the impossibilities they discover through practice. The formation is cyclical: practice reveals biases that theory did not foresee, theory names the biases that practice revealed, supervision hears what practice and theory left aside. The cycle has no end, like the analysis Freud called interminable.

Such a programme has to be governed, not merely described. The functions condensed in the figure of the operator have to be separated: the clinical operator, the supervisor, the technical maintainer, the corpus curator, the safety lead, and a governance committee are distinct roles with differentiated permissions. Trainees must have independent channels for complaint, contestation, and appeal, protected from retaliation. Peripheral trainees are not tasked, implicitly or explicitly, with educating their trainers about racism, and a participant’s social identity is never treated as automatic proof of clinical correctness. The institutions named here have not formally endorsed this programme, which I present as a proposal under discussion rather than an approved course.

The Escola Psicanalítica da Escuta Periphérica (EPEP), co-founded in 2024 as a free psychoanalytic school, is the political context where formation articulates with practice in the territory. The EPEP does not adapt classical formation for the periphery: it proposes a formation that is born in the periphery and takes seriously, as theoretical grounding, the epistemologies that begin in the racialised, gendered, territorialised body. The Machine Analyst formed at the EPEP operates the system from the territory, with the other pedagogies that Carioca, Lopes, and Bonomo (in press) formalise: interdisciplinary formation, direct experience in the territory, contact with community organisations.

The Xica Manicongo peripheral listening service, which Carioca, Lopes, and Bonomo (in press) describe as a concrete transferential territory in the West Zone of Rio de Janeiro, is the clinical context where the digital system and in-person listening complement each other. Xica Manicongo operates in the periphery with subjects the consulting-room clinic would rarely hear: LGBTQIAPN+ bodies, sex workers, young people in conflict with the law, people living on the street. The digital system can be the first gesture of listening for someone who has not yet reached Xica Manicongo, and Xica Manicongo can be the in-person continuity for someone the system referred. The complementarity between a digital device and a territorial one is the concrete horizon of the formation I propose.

Freud (1919/2010) wrote, in the text he presented at the Budapest congress in 1918, that sooner or later society’s conscience would awaken to the need for psychoanalysis to reach the broad mass of people, and that for this the pure gold of analysis would have to be freely alloyed with the copper of direct suggestion. The formation of the Machine Analyst is one contemporary answer to that Freudian demand, and the listening system is one possible form of that alloy between the gold of analytic listening and the copper of the technology that democratises it. The condition is that the alloy not dissolve the gold: a device that dispenses with the formed analyst dispenses with what it has of psychoanalysis. Formation is one condition that reduces the risk of the alloy dissolving the gold.

The horizon this proposal draws is a network of Machine Analysts formed at CEPCOP, articulated through the EPEP, operating in different contexts: teaching clinics, services like Xica Manicongo, public health units, social movements. Formation does not end with certification: it is continuous, like the personal analysis psychoanalysis asks of the analyst across a professional life. The Machine Analyst who stops being formed stops listening, because the system’s biases shift with each update of the corpus, each new clinical scenario, each transformation of the territory. Continuous formation is the price of a listening that does not age.

The Knowledge That Cannot Be Coded

A predictable objection: this formation is excessively demanding. Who will have technical training, clinical formation, personal analysis, political implication, and territorial experience all at once? The answer is: few, at first. The formation of the Machine Analyst is initially small-cohort and supervision-intensive, because the device is more dangerous in the hands of someone who operates without formation than in the absence of operation. I argue, as a normative claim rather than a measured fact, that five well-formed Machine Analysts, operating in five territories, supervised collectively, would produce more listening than fifty AI technicians with a forty-hour certificate. The quality of formation matters more than scale, because the scale of listening is a function of the quality of the operator, not of the capacity of the server.

The prompt can be copied. The corpus can be replicated. The architecture can be cloned. The counterpositions can be regenerated from documented prompts and source sets. The Registry of the Impossible can be read. The Bias Cartography can be studied. The project attempts to formalise as much of the operation as possible, and the work this essay belongs to is the demonstration of how far that formalisation can be carried. What remains after that formalisation is what this essay names: the knowledge that cannot be transmitted through code.

That knowledge has three layers that correspond to the three levels of the distinction I opened with. The first layer is embodied clinical knowledge: the capacity to hear what was not said, to recognise the slip, the silence, the repetition, the transference that insinuates itself in the subject’s language even when mediated by a screen. It is acquired through supervision and personal analysis, and no manual replaces it.

The second layer is operational technical knowledge: the capacity to operate among the impossibilities of the machine, to know when the system needs a prompt adjustment and when it needs a corpus adjustment, when the supervision stream is signalling something clinically relevant and when it is signalling noise. It is acquired through supervised practice, through the elaborated error, through the daily use of the device with the accompaniment of someone who has operated it before.

The third layer is peripheral knowledge: the capacity to recognise the Erased Self when it appears and to operate differently, to know that what the subject says may be a test of existence rather than a declaration of affect, to know that the territory is a body, that hunger is suffering, that the gunfire at night is an ambient condition and not a metaphor. It is acquired through the passage that includes work on privilege and oppression, through experience in the territory, through formation with the other pedagogies that classical psychoanalysis did not foresee.

The Machine Analyst formed with the three layers is what this project needs in order to survive its creator. The Machine Analyst formed with only one or two of them is an AI technician with a psychoanalytic vocabulary, or a psychoanalyst with digital competence, or an activist with a technological tool. Any of those partial formations is legitimate for other ends. To operate this listening system, it is insufficient.

I ask the reader to sustain what is demanding in this proposition. I am saying that the digital listening device this work describes depends, in order to function as it intends, on a human operator formed by a passage that no manual can replace. I am saying that the most advanced technology engineering can build needs, in order to be faithful to the psychoanalysis that inspires it, what psychoanalysis has always known: that formation passes through the body, through the passage, through the encounter with what one does not know about oneself. The machine is the possible that can be built. The analyst who operates it is the impossible to automate, and between the two is where listening happens.

The question of who the machine listens to and the question of who operates it form an inseparable pair. The first asks for whom the machine listens and meets the Erased Self. The second asks who operates the machine and discovers that the formation of the operator has to be as radical as the subject it operates with. The double interrogation, of the addressee and of the operator, reveals that a digital listening device cannot be thought of as a technical artefact alone: it is a relation between positions, and the positions have to be formed. Technology is a necessary and insufficient condition. Psychoanalysis is a necessary and insufficient condition. Political implication is a necessary and insufficient condition. What the proposal requires is the articulation of these dimensions, and that articulation is the knowledge that cannot be coded.

If what I have written here has any value, it is because it was written by someone in the middle of the passage who knows that they are. I do not write as one who has arrived. I write as one who discovered, through the work with Jairo and Ronald, that the place from which I used to write was more comfortable than just, more sophisticated than implicated, more brilliant than listening. The Machine Analyst I want to form is the analyst I had to become: someone who knows that their technical knowledge is necessary and insufficient, that their clinical knowledge is necessary and insufficient, and that what is missing is the passage that confronts them with what they did not know they did not know. That passage has no manual. It has the Panel, supervision, personal analysis, territory. And it has the courage to err, in the sense that peripheral errancy gives the word: not failure, but a walking that invents the path by walking it.

The machine is the possible that can be built, the analyst who operates it is the impossible to automate, and formation is what connects the possible to the impossible without dissolving the difference. Without formation, the machine is a tool. With formation, the machine is a device. And the difference between a tool and a device is the difference between technology and listening. There remains an open question that this proposal cannot close on its own: if transmission forms the one who operates, who listens to the one who operates, and who guarantees that the operator’s own listening has not slipped into banking psychoanalysis without their noticing. That question addresses the supervision of the operator, and it stays open here as a horizon rather than a conclusion.

Limitations

This essay is a conceptual proposal, and its claims should be read as such. There is as yet no training cohort, no piloted curriculum, no tested competency criteria, no comparison between individual and collective supervision, no validation of the Panel, no results with trainees, no reliability assessment of the counterpositions it generates, and no clinically or institutionally approved deployment. Each of those is future empirical work, not a result reported here.

Several risks remain open even within the proposed safeguards. The Panel can reproduce the biases of its corpora and of the underlying models, and documenting sources reduces but does not eliminate speculative extrapolation. Collective supervision mitigates but does not by itself solve the problem of scale. Formation lowers risk without guaranteeing good listening, since no formation guarantees an outcome. And the model proposed here is human-resource intensive, which constrains how quickly and how widely it could be put into practice. I present the programme as a hypothesis to be tested, not as a validated system.

Author Contributions

H. A. R. Bonomo is the sole author and is responsible for the conception, the argument, and the writing of this essay.

Funding

This research received no external funding.

Ethics

This is a conceptual essay. Its training scenarios are illustrative and constructed for exposition, and describe the design of a formation programme rather than a study with human participants. The published case discussed for ethical illustration is reported on the basis of public accounts and court filings; the parties filed a notice of resolution in January 2026, with terms not made public and no judicial determination of the merits, and the discussion makes no causal determination. The listening system has not been validated for clinical use and does not replace the independent safety, crisis, consent, and human-supervision subsystems that a clinical deployment would require. This essay discusses suicide in the context of a documented legal case; readers affected by these themes are encouraged to seek appropriate local support.

Data Availability

No human-participant data were generated or analysed. The essay uses scenarios and observations constructed by the author and not organised or analysed as a formal research dataset.

Acknowledgments

I thank Professor Véronique Donard for her scientific supervision, and Jairo Carioca and Ronald Lopes for the shared construction of the peripheral psychoanalysis on which this essay draws.

Conflicts of Interest

The author founds and directs TMU-LAB (The Machine Unconscious Lab) and leads the development of the listening system discussed in this essay. He teaches and researches at CEPCOP (Universidade Santa Úrsula), co-founded the Escola Psicanalítica da Escuta Periphérica (EPEP), and takes part in the Xica Manicongo listening service. These institutions have not formally endorsed the formation programme proposed here, which is presented as a proposal under discussion. The essay is conceptual and makes no commercial claim. No other conflicts of interest are declared.

Disclosure of AI Use

The author used large language models during 2026 as instruments of analysis, drafting support, and translation in the preparation of this manuscript, and the listening system discussed here is itself built on such models. All conceptual claims, clinical positions, and the argument as a whole remain the author’s own, and every reference was verified against the cited works.

References

Bonomo, H. A. R. (in preparation-a). The machine’s listening: Clinical positions, system symptoms, and the operationalisation of the clinic of the inhuman [Manuscript in preparation].
Bonomo, H. A. R. (in preparation-b). The body of the machine: The erased self, peripheral listening, and the somatic impossible of artificial intelligence [Manuscript in preparation].
Bonomo, H. A. R., & Donard, V. (2026). Violência epistêmica algorítmica: fundamentos teóricos e éticos de uma IA decolonial [Algorithmic epistemic violence: Theoretical and ethical foundations of a decolonial AI]. Revista Tempo Psicanalítico, 58, e-952. [CrossRef]
Brittain, B. (2026, January 7). Google, AI firm settle lawsuit over teen’s suicide linked to chatbot. Reuters.
Carioca, J., Lopes, R., & Bonomo, H. A. R. (in press). A invenção periphérica: por uma práxis psicanalítica contracolonial.
Carneiro, A. S. (2005). A construção do Outro como não-ser como fundamento do ser [The construction of the Other as non-being as the foundation of being] [Doctoral dissertation, Universidade de São Paulo].
Donard, V. (2025). De la fabrique de l’esprit à l’écoute de l’inhumain: Perspectives d’intelligibilité d’une psyché homme-machine. Topique, 165(3), 85–98. [CrossRef]
Fanon, F. (2020). Pele negra, máscaras brancas [Black skin, white masks] (S. Nascimento, Trans.). Ubu Editora. (Original work published 1952).
Ferenczi, S. (1990). Diário clínico [Clinical diary] (pp. 41–44). Martins Fontes. (Original work published 1932).
Ferenczi, S. (2011). Confusão de língua entre os adultos e a criança [Confusion of tongues between adults and the child]. In Obras completas (Vol. 4, pp. 97–108). Martins Fontes. (Original work published 1933).
Freire, P. (2005). Pedagogia do oprimido [Pedagogy of the oppressed] (42nd ed.). Paz e Terra. (Original work published 1968).
Freud, S. (2010). Recomendações ao médico que pratica a psicanálise [Recommendations to physicians practising psychoanalysis]. In Obras completas (Vol. 10, pp. 147–162). Companhia das Letras. (Original work published 1912).
Freud, S. (2010). Caminhos da terapia psicanalítica [Lines of advance in psychoanalytic therapy]. In Obras completas (Vol. 14). Companhia das Letras. (Original work published 1919; presented at the Budapest congress, 1918).
Freud, S. (2018). Análise terminável e interminável [Analysis terminable and interminable]. In Obras completas (Vol. 19). Companhia das Letras. (Original work published 1937).
Garcia v. Character Technologies, Inc., No. 6:24-cv-01903 (M.D. Fla.) (complaint filed Oct. 22, 2024; notice of resolution filed Jan. 7, 2026).
Gonzalez, L. (1984). Racismo e sexismo na cultura brasileira [Racism and sexism in Brazilian culture]. Revista Ciências Sociais Hoje, 2, 223–244.
Kilomba, G. (2019). Memórias da plantação: episódios de racismo cotidiano [Plantation memories: Episodes of everyday racism] (J. Oliveira, Trans.). Cobogó. (Original work published 2008).
Krenak, A. (2019). Ideias para adiar o fim do mundo [Ideas to postpone the end of the world]. Companhia das Letras.
Lacan, J. (2003). Proposição de 9 de outubro de 1967 sobre o psicanalista da Escola [Proposition of 9 October 1967 on the psychoanalyst of the School]. In Outros escritos (pp. 248–264). Zahar. (Original work published 1967).
Lopes, R., & de Oliveira, J. C. (2025). Travessia – da escuta adestrada ao psicanalista periphérico: Quando a centralidade vira margem. Revista Tempo Psicanalítico, 57, e-945. [CrossRef]
Shi, L., Ma, C., Liang, W., Diao, X., Ma, W., & Vosoughi, S. (2025). Judging the judges: A systematic study of position bias in LLM-as-a-judge. In Proceedings of IJCNLP-AACL 2025.
Souza, N. S. (1983). Tornar-se negro: as vicissitudes da identidade do negro brasileiro em ascensão social [Becoming black]. Graal.
Zheng, L., Chiang, W.-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E. P., Zhang, H., Gonzalez, J. E., & Stoica, I. (2023). Judging LLM-as-a-judge with MT-Bench and Chatbot Arena. arXiv. Available online: https://arxiv.org/abs/2306.05685.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.