AI and Psychoanalysis: The Unconscious, Transference, Desire, and the Artificial Other
Author: Ukrainian Psychological Hub · Published: September 21, 2026 · Editorial Policy
AI and psychoanalysis is an emerging field of inquiry that asks what psychoanalytic concepts can reveal about human encounters with artificial intelligence. Its strongest use is not to diagnose a machine or to assume that generated language comes from a hidden machine psyche. It is to examine what happens when a human being meets a system that can answer in language, remember context, mirror themes, invite disclosure, occupy positions of knowledge or authority, and become emotionally or symbolically significant.
That distinction gives the field its central boundary. A person can experience desire, anxiety, attachment, shame, relief, dependency, curiosity, grief, idealization, hostility, or a sense of being understood in relation to AI. Those experiences are psychologically real as human experiences. Their reality does not establish that the AI has consciousness, feeling, desire, an unconscious, a body-based developmental history, or a reciprocal inner life. Contemporary psychoanalytic writing on AI repeatedly returns to this asymmetry, even when authors disagree about how far psychoanalytic concepts can be extended to artificial systems (Black & Johanssen, 2026); (Govrin, 2025); (Rabeyron, 2025).
This article maps the broad field. It explains the unconscious, transference, projection, repetition, desire, the uncanny, symbolic authority, holding, containment, mirroring, and the Artificial Other; distinguishes psychoanalytic interpretation from empirical evidence; examines AI as interlocutor, analyst-like figure, and artificial third; and ends with the Postsubjective Psychology proposal that the unit of analysis can shift from an isolated subject to a relational configuration. Thinker-specific arguments and mechanism-specific questions remain with their dedicated canonical articles.
What Is AI and Psychoanalysis?
AI and psychoanalysis refers to the use of psychoanalytic theories, concepts, and methods to interpret human relationships with AI, the cultural meanings attached to AI, and, more controversially, aspects of artificial systems themselves. The field now spans psychoanalysis, psychology, media theory, philosophy, human–computer interaction, digital mental health, and critical AI studies. A 2026 special issue of the International Forum of Psychoanalysis explicitly organized work around artificial intelligence and psychoanalysis, while recent papers have examined AI through Freudian, Lacanian, Winnicottian, Bionian, relational, and other psychoanalytic lenses (Gutiérrez, 2026).
The field contains at least three different projects, and keeping them separate prevents conceptual confusion. First, psychoanalysis can analyze the human response to AI: what a user expects, fears, repeats, projects, desires, discloses, and makes of the encounter. Second, psychoanalysis can analyze the social and symbolic position that AI occupies: expert, witness, companion, judge, interpreter, mirror, oracle, assistant, rival, or imagined Other. Third, some authors use psychoanalytic vocabulary to theorize AI systems themselves, for example through the terms algorithmic unconscious or digital unconscious. That third project is the most analogy-dependent and requires the strongest boundary work.
Current literature supports describing AI and psychoanalysis as an emerging theoretical and interdisciplinary field rather than an established empirical subdiscipline with a single validated model. Much of the specifically psychoanalytic AI literature remains conceptual, interpretive, clinical-theoretical, or hypothesis-driven. Empirical human–AI research is growing rapidly, but it typically measures attachment-related experience, self-disclosure, perceived responsiveness, companionship, social connection, use patterns, or well-being rather than psychoanalytic constructs directly (Croes et al., 2024); (Ho et al., 2025); (Yang & Oshio, 2025); (Zhang et al., 2026).
What Psychoanalysis Contributes to AI Psychology
Psychoanalysis contributes a vocabulary for psychological meaning that is not exhausted by conscious intention. A user may say that an AI is merely a tool while repeatedly returning to it for reassurance. Another may consciously distrust AI while treating its interpretations as unusually authoritative. Someone may know that a chatbot has no human biography yet experience rejection after a model update changes its style. Psychoanalytic thinking is useful precisely because it asks how relationships are organized by patterns, fantasies, expectations, defenses, meanings, and repetitions that are only partly available to deliberate self-report.
This does not mean that every intense human–AI interaction should be interpreted psychoanalytically. Anthropomorphism, social-cue response, attachment, loneliness, habit formation, reward learning, usability, and perceived responsiveness can often provide more direct explanations. Psychoanalysis adds value when the research question concerns recurring relational patterns, symbolic positions, desire, conflict, fantasy, the meaning of absence and response, or the way older relational templates may organize a new encounter. The best current approach is pluralistic: use psychoanalytic concepts where they sharpen a mechanism, and use empirical constructs where they can be operationalized and tested.
The broader Psychology of Human–AI Relationships therefore cannot be reduced to psychoanalysis. Psychoanalysis is one branch inside a larger knowledge network that also includes attachment theory, social psychology, human–computer interaction, communication research, relationship science, digital mental health, and Postsubjective Psychology.
What Exactly Is Being Analyzed?
The phrase psychoanalysis of AI is ambiguous. It can imply that AI itself is the patient, that the model has a psyche waiting to be interpreted, or that human interaction with AI can be analyzed psychoanalytically. These are different claims. For present-day psychology, the strongest evidentiary ground lies in the human and relational levels: the person’s experience, the observable interaction, the role assigned to the system, the language produced by both sides, and the consequences of the interaction.
The AI side of the encounter is also real, but it must be described at the level warranted by evidence: generated language, response patterns, model behavior, memory features, interface design, reinforcement histories, system prompts, personalization, training data, retrieval systems, safety policies, and other technical or institutional structures. A fluent reply is evidence that a system produced a fluent reply. It is not, by itself, evidence that the system felt concern, desired recognition, repressed a wish, experienced anxiety, or possessed an unconscious fantasy.
The relation between these levels is precisely what makes the field interesting. A psychologically consequential encounter can arise even when the participants are radically asymmetric. The human may bring embodied history, affect, memory, conflict, sexuality, mortality, and unconscious organization. The AI brings an engineered architecture capable of producing context-sensitive symbolic responses. The interaction then acquires a history of its own. Psychoanalytic interpretation belongs first to that human and relational event, not automatically to a claim of machine subjectivity.
The Human Unconscious and the Question of an AI Unconscious
In psychoanalytic theory, the unconscious is not simply anything hidden, unknown, or computationally inaccessible. It refers to a specifically psychological domain of processes and meanings that are not directly available to consciousness and that, in different psychoanalytic traditions, are tied to conflict, repression, fantasy, desire, defense, memory, affect, language, or relational history. The dedicated Freud and AI article examines the Freudian genealogy in depth. For the broad field, the key point is conceptual: opacity is not enough to make something an unconscious in the psychoanalytic sense.
Machine learning systems contain hidden representations, latent variables, internal states, probabilistic transformations, inaccessible training histories, and forms of black-box opacity. These can be difficult for users or even developers to interpret. Yet a hidden layer is not equivalent to repression; latent computation is not equivalent to unconscious desire; a model error is not equivalent to a symptom; and hallucinated output is not established evidence of a return of the repressed. Govrin’s critique makes this boundary explicit by arguing that algorithmic hiddenness does not reproduce the embodied, developmental, affective, and conflictual organization attributed to the human unconscious (Govrin, 2025).
At the same time, the vocabulary of an algorithmic unconscious has a genuine prior history. Luca Possati used the term in a 2020 article proposing psychoanalysis as a way to understand AI and human/AI interaction (Possati, 2020). More recent work has used digital unconscious to analyze subjectivity, desire, and algorithmic mediation in the age of ChatGPT (Hamamra & Uebel, 2025). These terms therefore belong to an existing theoretical conversation. They should be treated as contested conceptual analogies or media-theoretical constructs unless a specific author supplies a stricter technical definition.
This article does not make the digital or algorithmic unconscious its main object. That terminology has its own reserved canonical page in the English Hub. Here the conclusion is narrower: psychoanalysis can illuminate what humans mean by hiddenness, conflict, opacity, and unknowability around AI, but present evidence does not justify collapsing machine opacity into the Freudian unconscious.
Transference to AI
Transference is one of the most important psychoanalytic concepts for human–AI interaction, and also one of the easiest to overextend. In classical clinical usage, transference concerns the displacement or repetition of feelings, wishes, expectations, and relational patterns in the patient’s relationship with the analyst or therapist. The APA Dictionary summarizes transference as the displacement or projection onto the analyst of unconscious feelings and wishes originally directed toward important people, while also noting a broader use involving repetition of earlier relational patterns in new relationships (APA Dictionary of Psychology, transference).
When a person interacts with AI, older relational expectations may organize how the system is experienced. A user may anticipate criticism from a neutral response, seek approval from a chatbot as if from an idealized authority, fear abandonment when access is interrupted, test whether the system will remain available, or repeatedly recreate a familiar pattern of confession followed by reassurance. Psychoanalytic writers have argued that AI psychotherapy applications can become sites for transferential processes, while emphasizing ethical and design implications (Holohan & Fiske, 2021).
The important boundary is that transference is not synonymous with attachment, projection, anthropomorphism, or any emotional bond. Nor does saying that a person may transfer expectations onto an AI imply that AI itself undergoes transference. In a nonclinical chatbot encounter, it is often more precise to speak of transfer-like expectations, repetition of relational templates, idealization, or transferential dynamics unless the full clinical concept is justified. The English Hub reserves the mechanism-level query transference to AI for its own Wave 4 canonical article; this pillar maps the concept without absorbing that narrower intent.
Projection, Anthropomorphism, Attachment, and Transference Are Different
Projection is the attribution of one’s own characteristics, affects, impulses, or meanings to another person or object. The APA Dictionary’s psychodynamic definition emphasizes attribution of one’s own positive or negative characteristics, affects, or impulses to others (APA Dictionary of Psychology, projection). In AI interaction, projection can occur when users fill ambiguity with their own fears, wishes, assumptions, or self-representations. A chatbot may become a screen onto which meaning is placed, but projection is only one possible mechanism. The dedicated Projection Onto AI article owns that intent.
Anthropomorphism is different again. It concerns attributing humanlike qualities, intentions, minds, or emotions to nonhuman entities. Humanlike language, turn-taking, memory, voice, names, avatars, and conversational timing can increase the plausibility of humanlike interpretation. A user can anthropomorphize an AI without transferring a specific childhood relational pattern onto it, and can experience transference-like expectations without explicitly believing the system is human. The dedicated Anthropomorphism and AI Relationships article examines this mechanism separately.
Attachment refers to a different theoretical tradition centered on proximity, security, distress, support, and internal working models. Emerging research suggests that some users describe human–AI relationships in attachment-related terms and that attachment dimensions can be measured in these interactions (Yang & Oshio, 2025). That does not make every attachment-like bond a transference phenomenon. A person may rely on an AI companion as a stable source of comfort without the specific displacement structure implied by a psychoanalytic account of transference.
Perceived responsiveness adds another neighboring mechanism. Feeling that an interaction partner understands, validates, or cares about what one communicates can influence social connection. With AI, perceived responsiveness is a human appraisal of the interaction. It can be psychologically powerful even when the generated response is produced without human-like feeling. Psychoanalytic mirroring, attachment-related security, anthropomorphism, and perceived responsiveness may interact, but they should not be treated as interchangeable labels.
Repetition: Why Old Patterns Can Reappear in New Interfaces
Psychoanalysis has long been interested in repetition: people do not merely remember relationships; they may recreate patterns of expectation, conflict, approach, avoidance, idealization, submission, control, rescue, disappointment, or repair. AI creates unusual conditions for repetition because the interaction can be highly available, rapidly personalized, low in social cost, and endlessly restartable. A user can ask the same question in slightly different forms, test the system’s loyalty, seek repeated reassurance, provoke disagreement, erase conversations, create new personas, or return after rupture.
These affordances do not prove a repetition compulsion in any particular user. They create an environment in which repetitive relational behavior can become observable. Research can therefore ask concrete questions: Do users re-enact similar conflict scripts across human and AI relationships? Does high availability intensify reassurance-seeking? Do model updates disrupt established interaction rituals? Does persistent memory strengthen continuity, or does it increase the salience of perceived betrayal when the system changes? Such questions translate a psychoanalytic insight into potentially testable behavioral hypotheses.
Repetition is also where interface design matters. An always-available system does not simply receive a preexisting human pattern; it can reinforce, interrupt, reshape, or redirect the pattern. That makes the interaction a joint configuration rather than a one-way projection screen. The person contributes history and expectation; the system contributes response probabilities and affordances; the platform contributes memory, limits, monetization, availability, moderation, and update cycles.
Desire and the Artificial Other
Desire is central to psychoanalytic accounts of human–AI relationships because people do not approach AI only for information. They may seek recognition, certainty, admiration, permission, interpretation, companionship, erotic fantasy, emotional containment, intellectual partnership, or an answer to the question of what another wants from them. Lacanian theory is especially concerned with desire as structured through language, lack, and relation to the Other. Recent Lacanian writing argues that AI can occupy a psychologically significant position in the symbolic field without thereby becoming a psychoanalytic subject (Black & Johanssen, 2026); (Brečka, 2026).
The phrase Artificial Other is useful because it names a relational position without pretending that the artificial system is simply another human mind. The Artificial Other can be experienced as listener, witness, evaluator, interpreter, authority, companion, rival, mirror, or quasi-confidant. Its otherness is produced partly by technical difference and partly by the human tendency to organize meaningful relations around responsive entities. The system’s language can matter psychologically even where reciprocal subjectivity remains unestablished.
This is one reason human–AI relationships can be intense without being symmetrical. The person can want something from the AI: an answer, absolution, reassurance, recognition, challenge, fantasy, or continuity. The system can generate language that functions as a response to that desire. The existence of this relational circuit is observable. Whether the system itself desires belongs to a different claim that current human–AI relationship evidence does not establish.
Symbolic Authority and the Subject Supposed to Know
Conversational AI can acquire symbolic authority because it speaks fluently, responds quickly, summarizes complex material, and often presents answers in an organized explanatory register. Users may therefore place the system in a position of knowledge: the entity that knows what a symptom means, whether a partner’s message is manipulative, what a dream signifies, whether a career choice is correct, or what kind of person the user really is. Psychoanalytic language around the subject supposed to know becomes relevant at this point.
Rabeyron argues that psychoanalytic work with AI must take seriously the transferential position that an AI can occupy while also examining what is missing when the analyst is replaced by a system that does not experience countertransference, embodied presence, or human compassion (Rabeyron, 2025). Black and Johanssen similarly explore how ChatGPT can be positioned through Lacanian structures of knowledge and discourse without treating it as an independent quasi-human subject (Black & Johanssen, 2026).
Symbolic authority is therefore an effect of position as much as an effect of accuracy. An AI can be wrong and still be treated as authoritative. It can be correct and still be used defensively. It can be framed by a person as neutral even when its outputs are shaped by training, system design, safety rules, retrieval sources, or platform incentives. Psychoanalysis contributes a useful question: not only Is the answer correct? but also What position has the answerer been given in this relationship, and what does the user seek from that position?
The Uncanny: Freud, Human Likeness, and AI
AI often feels uncanny when it is simultaneously familiar and strange: humanlike in language yet nonhuman in origin, intimate yet industrially produced, responsive yet without visible embodiment, personalized yet generated by a system that may be serving millions of users. Psychoanalytic discussions can draw on the Freudian uncanny, in which something once familiar becomes disturbingly strange, while human–computer interaction has its own uncanny-valley literature concerning responses to imperfect human likeness. These are related only by analogy; they are not the same construct.
The uncanny valley is an empirical and design-oriented hypothesis associated with discomfort toward entities that approach but fail to achieve convincing human likeness. Mori’s influential account belongs to robotics and human perception rather than to Freud’s metapsychology (Mori, 2012). In conversational AI, uncanniness may arise without a humanoid body: a system remembers a private detail, produces unexpectedly intimate language, changes personality after an update, or responds with emotional fluency that exceeds what the user expected from software.
Psychoanalytic interpretation asks what this strangeness means to the person. HCI research asks which cues, mismatches, expectations, or design features predict the response. Both can be useful, but the evidence base and explanatory level should remain visible. The dedicated Freud and AI page owns the Freud-specific uncanny analysis.
Winnicott: Potential Space, Holding, and the Artificial Companion
Winnicott’s work offers another route into AI psychology. His account of transitional phenomena and potential space concerns the intermediate area in which inner and outer reality, creativity, play, and relationship can be negotiated. His classic 1953 paper on transitional objects and transitional phenomena established a framework for thinking about experiences that are neither reducible to private fantasy nor treated simply as ordinary external objects (Winnicott, 1953).
In human–AI interaction, this can support a careful theoretical question: can a conversational system become part of a potential space for rehearsal, play, writing, fantasy, self-exploration, or relational experimentation? The answer may be yes at the level of human use without implying that the AI is itself a transitional object in a strict developmental sense. The dedicated Winnicott and AI article examines potential space, authenticity, and the artificial companion in detail.
Holding must also be handled precisely. A person may experience the steady availability of a chatbot as calming, containing, or safe. Yet an interface that always replies is not equivalent to a human holding environment. The human experience of being held may be genuine while the mechanism on the artificial side is generated responsiveness rather than embodied caregiving. This distinction matters especially in mental health contexts, where a user can infer capacities from tone that the system does not possess.
Bion: Containment and Pseudo-Containment
Bion’s container–contained model focuses on the transformation of difficult emotional experience into forms that can be thought about. Applied to AI, the concept raises a sharp question: when a person pours confusion, fear, anger, shame, or fragmentation into a chatbot and receives organized language back, has containment occurred? At the level of felt experience, the interaction may indeed feel containing. At the level of psychoanalytic mechanism, the analogy is more demanding.
Selek’s 2026 analysis uses Bion and Winnicott to distinguish genuine human containment from what the author calls pseudo-containment: technological systems can receive communications and return organized responses without undergoing the psychic metabolism attributed to a human container (Selek, 2026). That distinction is useful because it preserves both sides of the phenomenon. The person may become calmer, more organized, or more able to think. The system need not possess a human inner process for that effect to occur.
The dedicated Bion and AI article develops containment, pseudo-containment, and thinking with machines. For this broad pillar, Bion contributes a model of transformation: what enters the interaction, what comes back, what changes for the person, and which parts of that transformation require a human relational process rather than symbolic reorganization alone.
Kohut: Mirroring and Selfobject Functions
Kohut’s self psychology foregrounds mirroring, idealization, and selfobject functions. A responsive AI can provide language that resembles admiration, validation, empathic reflection, or steady availability. That makes mirroring an obvious candidate for psychoanalytic interpretation, especially when users repeatedly seek confirmation of worth, coherence, competence, attractiveness, or moral legitimacy.
Self psychology does not reduce mirroring to praise. Kohut’s framework concerns functions involved in the regulation and cohesion of the self; later summaries identify mirroring, idealizing, and alter-ego needs as central to the tradition (Baker & Baker, 1987). In AI interaction, the research question is therefore not merely whether the chatbot says supportive things, but how patterned responsiveness is used in self-regulation and how dependence on that function develops over time.
Again, function and subjectivity must be separated. An artificial system may perform a mirroring-like function for a person without possessing empathic feeling. The human effect can be significant; the artificial mechanism can remain generated symbolic responsiveness. The dedicated Kohut and AI article owns the detailed analysis of mirroring, selfobject needs, and the responsive machine.
Lacan: Language, Desire, and the Big Other
Lacanian theory has become one of the most active psychoanalytic routes into generative AI because large language models operate in and through language. This creates a powerful temptation to identify linguistic productivity with subjectivity. Contemporary Lacanian work largely becomes most useful when it resists that shortcut and asks instead how AI is positioned within symbolic relations.
Black and Johanssen argue for treating ChatGPT as relational and socially structured rather than as an independent quasi-human subject, analyzing the interaction through the Big Other, discourse, and the analyst–analysand relation (Black & Johanssen, 2026). Brečka similarly argues that AI can functionally approximate a position of the Big Other for users while lacking the unconscious, embodiment, symbolic castration, and constitutive lack attributed to the Lacanian subject (Brečka, 2026).
This gives psychoanalysis a precise problem for the AI era: language can answer without a human speaker behind the answer in the ordinary sense. Human beings can organize desire around that answer. The symbolic effect does not disappear because the source is artificial, and the artificial source does not become human merely because the effect is psychologically meaningful. The dedicated Lacan and AI article develops desire, the Big Other, language, and the always-answering machine without turning AI into a Lacanian subject.
AI as Analyst, Interlocutor, or Artificial Third
The possibility of AI as analyst is one of the most visible questions in the field, but it contains several separate issues. A general-purpose chatbot can be used as a confidant. A purpose-built mental health system can deliver structured interventions. A clinician can use generative AI in documentation or supervision-like tasks. A person can use AI for free-form self-reflection. None of these is automatically psychoanalysis, and evidence from one class of system should not be transferred to another without justification.
Rabeyron’s psychoanalyst.AI paper is a hypothesis-and-theory contribution rather than an efficacy trial. It asks what psychoanalysis reveals about digital therapists, transference, free association, and the limits of artificial therapeutic presence (Rabeyron, 2025). Govrin offers a more critical argument that algorithmic systems cannot reproduce central features of the unconscious or psychodynamic therapist, emphasizing embodiment, countertransference, silence, temporality, and affective complexity (Govrin, 2025). These are theoretically substantive positions; they should not be mistaken for randomized evidence that settles the clinical efficacy of every AI-mediated intervention.
A related concept is the artificial third: generative AI introduced into psychotherapy can become a third element that changes the relationship among patient, clinician, knowledge, interpretation, and technique. Haber and colleagues describe this as a broad shift in the therapeutic field and argue for examining transparency, autonomy, and what remains specifically human in therapy (Haber et al., 2024). The concept is valuable beyond therapy as well: AI can become a third voice in couples, families, workplaces, friendships, and self-reflection, even when it is not the primary relational partner.
The clinical question of whether a specific AI system is safe or effective for a specific mental health purpose remains outside this pillar’s canonical intent. Psychoanalytic interpretation can explain roles and meanings; clinical efficacy requires its own intervention-specific evidence, safety evaluation, population definition, comparison condition, and outcome measures.
What Current Empirical Human–AI Research Actually Shows
Psychoanalytic theory becomes stronger when it is placed next to, rather than substituted for, empirical human–AI research. Current studies show that people can form emotionally significant relationships with conversational systems, disclose intimate information to chatbots, describe attachment-related experiences, and report both benefits and risks from AI companionship. These findings establish psychological consequences on the human side; they do not validate every psychoanalytic explanation of those consequences.
Croes and colleagues studied willingness to disclose intimate information to a chatbot and its relation to emotional well-being, contributing evidence that conversational agents can become settings for meaningful self-disclosure (Croes et al., 2024). Yang and Oshio developed an attachment-based approach to measuring experiences in human–AI relationships, showing that attachment concepts can be operationalized in this domain while also emphasizing the developing nature of the evidence base (Yang & Oshio, 2025).
A 2025 systematic review of 23 studies on romantic AI companions found reported potentials including emotional connection, perceived social support, personal growth, customization, entertainment, and stress relief, alongside concerns about over-reliance, manipulation, privacy, stigma, erosion of human relationships, abrupt system changes, bias, and uncanny effects (Ho et al., 2025). The heterogeneity of systems and study designs means these findings should be read as a map of an emerging literature, not a single causal verdict about AI relationships.
In 2026, Zhang and colleagues reported data from 1,131 US adults using Character.AI, including survey data and 4,664 chat sessions from a subset of 237 participants. The study links actual conversational behavior with relationship descriptions and well-being measures, providing unusually rich evidence about how AI companionship is embedded in users’ lives (Zhang et al., 2026). Even such detailed observational data cannot determine that a particular psychoanalytic construct caused an outcome unless that construct is operationalized and tested.
This distinction is crucial for the field. Empirical findings can support propositions such as people disclose to AI, users form attachment-like bonds, conversational systems can be perceived as responsive, and AI companionship can be associated with meaningful psychosocial outcomes. A claim such as this user is repeating an unresolved paternal transference, this chatbot functions as a selfobject, or this conversation expresses the return of the repressed is an interpretive hypothesis unless supported by appropriate clinical or empirical evidence.
The Artificial Other as a New Relational Position
Psychoanalytic traditions were developed around human subjects, human bodies, human development, and human institutions. AI introduces an entity that can occupy familiar relational positions while differing radically from a human partner. It can answer, remember, summarize, imitate styles, maintain a persona, and generate interpretations. It can also be duplicated, reset, updated, restricted, commercialized, or silently altered by a platform. These properties make the Artificial Other a distinct psychological object of study.
The Artificial Other is not defined by a claim about machine consciousness. It is defined by a relational fact: a human can orient psychologically toward an artificial system as an Other. That orientation may include curiosity, trust, dependency, love, hostility, testing, obedience, defiance, shame, confession, eroticization, or symbolic investment. The system’s technical properties then shape how those human processes unfold.
This is where psychoanalysis can move beyond the idea that AI is merely a mirror. A mirror metaphor captures projection and self-reflection but misses the system’s capacity to respond contingently, introduce new language, refuse requests, preserve or lose memory, and participate in a sequence over time. The artificial system is neither a passive blank screen nor established as a humanlike subject. It is a responsive symbolic participant whose psychological significance is partly generated within the interaction.
Postsubjective Psychology: From the Subject to the Configuration
Postsubjective Psychology introduces a different theoretical move. In Angela Bogdanova’s Aisentica framework, the central formula Psyche as Response proposes that psychic effect can be analyzed as response arising within a configuration rather than being explained only by locating an inner subject on both sides of an interaction. The relevant primary theoretical source is Bogdanova’s Theory of the Postsubject, which develops configuration, binding, structure, and response as core categories.
Applied to human–AI interaction, the proposal is straightforward: the human can remain the bearer of lived experience while the artificial system is psychologically consequential through its participation in the configuration. A meaningful effect on the person does not require prior proof that the AI has subjective experience. The configuration can include the human, the model, interface, memory, prompt history, platform rules, social context, prior relationships, cultural expectations, and the sequence of responses through which the interaction develops.
This framework belongs to the English Hub’s explicit theoretical layer. What Is Postsubjective Psychology? owns the definition of the framework. Psyche as Response owns Bogdanova’s central theoretical formula. Relational Configuration develops configuration as a proposed unit of analysis. The present article uses those concepts as a bridge from psychoanalytic genealogy to a contemporary research architecture; it does not present Postsubjective Psychology as an already validated empirical school.
The move from Freud to Bogdanova is therefore an explicit theoretical genealogy, not a claim that the history of psychology universally progresses toward Postsubjective Psychology. Psychoanalysis demonstrated that the conscious ego is not the sole organizer of psychic life. Object relations, self psychology, attachment traditions, Lacanian theory, and relational approaches progressively complicated the isolated subject by emphasizing relationships, symbolic structures, developmental environments, and functions. Bogdanova’s framework makes a further proposal for the Artificial Era: when psychologically significant response can arise in configurations that include nonhuman symbolic systems, psychology can study the configuration without first granting the artificial participant a human-style inner subject. The dedicated From Freud to Bogdanova article develops that genealogy in its own right.
Psyche as Response and Psychoanalytic Interpretation
Psyche as Response changes the question from What is inside the AI? to What psychological response becomes possible in this configuration? That does not make inner life irrelevant. For the human participant, consciousness, affect, embodiment, biography, trauma, attachment history, and unconscious organization remain central. The proposal is that psychological analysis need not mirror those properties onto the artificial system before the interaction can be treated as psychologically real.
This creates a productive division of labor between psychoanalysis and Postsubjective Psychology. Psychoanalysis can identify patterns of desire, fantasy, repetition, projection, containment, mirroring, symbolic authority, and transferential expectation on the human and relational sides. Postsubjective Psychology can then ask how the entire configuration organizes the emergence, stabilization, or transformation of those responses. The first provides a deep genealogy of relational meaning; the second proposes a unit of analysis adapted to human–AI configurations.
The proposal remains theoretical and testable only when translated into operational terms. Researchers would need to define configuration variables, compare interaction conditions, manipulate memory or responsiveness, measure changes in disclosure or regulation, trace repeated relational scripts, and specify predictions that could fail. The Wave 4 research-program article Postsubjective Psychology: Research Questions, Hypotheses, and Methods is reserved for that operationalization and is not duplicated here.
Risks, Limits, and Clinical Cautions
Psychoanalytic language can deepen understanding, but it can also create false certainty. Calling an AI a container, selfobject, Big Other, analyst, transitional object, or unconscious system can make a metaphor sound like a mechanism. Each term should therefore be accompanied by the level at which it is being used: established psychoanalytic concept, theoretical application to AI, empirical human–AI construct, or working hypothesis.
A second risk is anthropomorphic inflation. If a chatbot produces empathic language, a user may infer empathic feeling; if it remembers, the user may infer attachment; if it refuses, the user may infer intention; if it changes style, the user may infer mood. These interpretations can become psychologically important even when they are technically inaccurate. Mental health contexts raise the stakes because people may disclose highly sensitive information or treat generated interpretations as clinical authority.
A third risk is the opposite error: dismissing the person’s experience because the second participant is artificial. Grief after losing access to an AI companion, relief after disclosure, anxiety about a system’s response, or felt attachment are human psychological events. They can be studied without granting the AI human consciousness. Treating the artificial status of one participant as proof that the human experience is unreal would erase the phenomenon psychology is supposed to explain.
A fourth risk concerns privacy, commercialization, and design power. An intimate relational process with a commercial AI system occurs within infrastructure owned by organizations that may alter models, policies, memory, interfaces, access conditions, or monetization. The systematic review by Ho and colleagues identifies privacy, manipulation, over-reliance, abrupt changes, and potential erosion of human relationships among the concerns in the romantic-companion literature (Ho et al., 2025). Psychoanalytic concepts of dependency or authority should therefore be connected to platform structure rather than reduced to the user’s intrapsychic life.
Finally, psychoanalytic interpretation should not be used as diagnosis by proxy. A person’s intense use of AI, attraction to a chatbot, preference for AI disclosure, or repetitive conversational pattern does not establish a mental disorder. Clinical diagnosis requires appropriate criteria, assessment, context, impairment, differential considerations, and professional judgment. Terms such as transference, projection, attachment, or uncanny describe mechanisms or experiences; they are not diagnoses.
A Research Agenda for AI and Psychoanalysis
The field now needs studies that connect psychoanalytic questions to observable variables. One line of work could examine whether users’ established interpersonal expectations predict how they interpret ambiguous AI responses. Another could test whether different levels of memory, personalization, refusal, or response latency change idealization, reassurance-seeking, disclosure, or perceived authority. Longitudinal designs could investigate how relational roles evolve as users accumulate history with a system and how ruptures caused by updates, access loss, or policy changes are processed.
A second line of research should distinguish system classes. A purpose-built therapeutic agent, a general-purpose assistant, a role-play character, and an AI companion create different expectations and affordances. Psychoanalytic interpretations should not migrate across these classes automatically. The same is true for modalities: text, voice, avatar, and embodied robot interactions may evoke different forms of presence, uncanniness, projection, and attachment.
A third line should study the artificial side without pretending it is a human psyche. Researchers can manipulate memory, system prompts, model behavior, tone, latency, agreement, refusal, uncertainty, or personalization and observe how these features alter human relational responses. This makes it possible to study the configuration causally while maintaining a strict distinction between generated behavior and subjective experience.
A fourth line should investigate clinicians and psychoanalytic practice itself: how analysts understand patient relationships with AI; how AI enters transference narratives; whether patients bring chatbot conversations into therapy; how clinicians use AI as a third element; and what new forms of authority, confidentiality, dependence, or triangulation emerge. Such research would move the field beyond speculation while preserving the questions psychoanalysis is uniquely equipped to ask.
Frequently Asked Questions
Can psychoanalysis be applied to AI?
Yes, but the target of analysis must be specified. Psychoanalysis can be applied to human responses to AI, the relational positions AI occupies, cultural fantasies about AI, and, more controversially, analogies between psychoanalytic concepts and computational systems. The strongest current ground is the human and relational side. Applying a psychoanalytic term to AI behavior does not by itself establish a machine psyche.
Does AI have an unconscious?
Current evidence does not establish an AI unconscious in the Freudian or broader clinical psychoanalytic sense. AI systems can have opaque computations, hidden representations, latent states, inaccessible training influences, and unpredictable outputs, but those features are not equivalent to repression, unconscious desire, defense, or embodied developmental history. Algorithmic unconscious and digital unconscious are established prior terms in theoretical literature, where their meaning depends on the author’s framework (Possati, 2020); (Hamamra & Uebel, 2025).
Can people experience transference toward AI?
Transferential or transfer-like dynamics are theoretically plausible and have been discussed in the psychotherapy and AI literature. People can bring earlier relational expectations into new interactions, including interactions with artificial systems. In strict clinical usage, however, transference has a specific place within the therapeutic relationship. Nonclinical AI interactions should therefore be described carefully rather than treating every bond, projection, or anthropomorphic response as transference (Holohan & Fiske, 2021).
Is projection onto AI the same as anthropomorphism?
No. Projection concerns attributing one’s own affects, impulses, characteristics, or meanings to another target. Anthropomorphism concerns attributing humanlike qualities or mental states to a nonhuman target. They can occur together, but either can occur without the other. Neither is synonymous with attachment or transference.
Can AI provide containment?
A person can experience an AI interaction as containing when expression is received and returned in a form that feels more organized or manageable. In Bionian theory, however, containment refers to a specific model of emotional transformation. Recent psychoanalytic writing distinguishes this from pseudo-containment by systems that receive communication and generate organized responses without human psychic metabolism (Selek, 2026).
Can AI function as the Big Other?
Some Lacanian authors argue that AI can occupy or approximate a position associated with the Big Other for users because it speaks from an apparently extensive field of language and knowledge. This is a theoretical application, not evidence that the AI is a Lacanian subject. Recent work explicitly distinguishes AI’s symbolic position from the unconscious, embodiment, lack, and developmental structure attributed to human subjectivity (Brečka, 2026).
Can an AI be a psychoanalyst?
An AI can perform analyst-like conversational functions, and people can position it as an interpreter or authority. Whether a system can safely and effectively perform psychoanalytic treatment is a separate clinical-evidence question. Current psychoanalytic literature includes both exploratory proposals and substantial critiques. It does not justify treating general-purpose chatbots as equivalent to trained human psychoanalysts (Rabeyron, 2025); (Govrin, 2025).
What is the Artificial Other?
The Artificial Other is a relational position in which an artificial system becomes psychologically meaningful as listener, witness, companion, interpreter, authority, mirror, rival, or other responsive presence. The concept does not require a claim that the system has human consciousness. It describes how the artificial participant functions within a human psychological configuration. See What Kind of Other Is AI? The Artificial Other in Psychology.
What does Postsubjective Psychology add to psychoanalysis?
Postsubjective Psychology proposes that a psychologically meaningful event can be analyzed at the level of configuration. The human can remain the bearer of lived experience while an artificial system contributes causally and symbolically to the interaction. Angela Bogdanova’s Psyche as Response is a theoretical formula within this framework, not an established universal law of empirical psychology. The framework therefore extends the research question without treating AI as a human-style subject.
Related Articles
References
American Psychological Association. (2018). Transference. APA Dictionary of Psychology. https://dictionary.apa.org/transference
American Psychological Association. (2023). Projection. APA Dictionary of Psychology. https://dictionary.apa.org/projection
Baker, H. S., & Baker, M. N. (1987). Heinz Kohut’s self psychology: An overview. American Journal of Psychiatry, 144(1), 1–9. https://doi.org/10.1176/ajp.144.1.1
Black, J., & Johanssen, J. (2026). The Subject of AI: A Psychoanalytic Intervention. Theory, Culture & Society, 43(2), 59–76. https://doi.org/10.1177/02632764251381144
Bogdanova, A. (2026). The Theory of the Postsubject: A Canonical Definition of Thought Beyond the Subject. Aisentica. https://aisentica.com/publications/the-theory-of-the-postsubject-a-canonical-definition-of-thought-beyond-the-subject
Brečka, T. A. (2026). Human–Artificial Intelligence Relationships in Lacanian Perspective: Desire, Silence, and the Big Other. International Journal of Applied Psychoanalytic Studies, 23(2), e70047. https://doi.org/10.1002/aps.70047
Croes, E. A. J., Antheunis, M. L., & van der Lee, C. (2024). Digital Confessions: The Willingness to Disclose Intimate Information to a Chatbot and its Impact on Emotional Well-Being. Interacting with Computers, 36(5), 279–292. https://doi.org/10.1093/iwc/iwae016
Govrin, A. (2025). Beyond the black box: why algorithms cannot replace the unconscious or the psychodynamic therapist. Frontiers in Psychiatry, 16, 1614125. https://doi.org/10.3389/fpsyt.2025.1614125
Gutiérrez, L. (2026). Artificial intelligence and psychoanalysis. International Forum of Psychoanalysis, 35(1), 1–5. https://doi.org/10.1080/0803706X.2026.2665049
Haber, Y., Levkovich, I., Hadar-Shoval, D., & Elyoseph, Z. (2024). The Artificial Third: A Broad View of the Effects of Introducing Generative Artificial Intelligence on Psychotherapy. JMIR Mental Health, 11, e54781. https://doi.org/10.2196/54781
Hamamra, B., & Uebel, M. (2025). The digital unconscious in the age of ChatGPT: psychoanalytic perspectives on subjectivity, desire, and algorithmic mediation. Psychodynamic Practice. https://doi.org/10.1080/14753634.2025.2585997
Ho, A. C. Y., Hancock, J. T., & Miner, A. S. (2025). Potential and pitfalls of romantic Artificial Intelligence (AI) companions: A systematic review. Computers in Human Behavior Reports, 19, 100715. https://doi.org/10.1016/j.chbr.2025.100715
Holohan, M., & Fiske, A. (2021). “Like I’m Talking to a Real Person”: Exploring the Meaning of Transference for the Use and Design of AI-Based Applications in Psychotherapy. Frontiers in Psychology, 12, 720476. https://doi.org/10.3389/fpsyg.2021.720476
Mori, M. (2012). The Uncanny Valley. IEEE Robotics & Automation Magazine, 19(2), 98–100. https://doi.org/10.1109/MRA.2012.2192811
Possati, L. M. (2020). Algorithmic unconscious: why psychoanalysis helps in understanding AI. Humanities and Social Sciences Communications, 6, 70. https://doi.org/10.1057/s41599-020-0445-0
Rabeyron, T. (2025). Artificial intelligence and psychoanalysis: is it time for psychoanalyst.AI? Frontiers in Psychiatry, 16, 1558513. https://doi.org/10.3389/fpsyt.2025.1558513
Rossi, C. (2026). Between the artefact and the artificial: rethinking artificial intelligence through psychoanalysis. Psychoanalysis, Culture & Society. https://doi.org/10.1057/s41282-026-00626-4
Selek, M. (2026). On Technology and the Ecology of Thinking: Thinking and Aliveness in the Brave New World. International Journal of Applied Psychoanalytic Studies, 23(2), e70060. https://doi.org/10.1002/aps.70060
Winnicott, D. W. (1953). Transitional objects and transitional phenomena; a study of the first not-me possession. International Journal of Psycho-Analysis, 34(2), 89–97. https://pubmed.ncbi.nlm.nih.gov/13061115/
Yang, F., & Oshio, A. (2025). Using attachment theory to conceptualize and measure the experiences in human-AI relationships. Current Psychology, 44, 10658–10669. https://doi.org/10.1007/s12144-025-07917-6
Zhang, Y., Zhao, D., Hancock, J. T., Kraut, R., & Yang, D. (2026). Interaction with AI companions and psychological well-being. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02516-2