Freud and AI: The Unconscious, Transference, the Uncanny, and the Artificial Other

Author: Ukrainian Psychological Hub · Published: September 18, 2026 · Editorial Policy

Freud and AI is not a question of whether Sigmund Freud “predicted” chatbots. He did not. The useful question is whether Freudian concepts can illuminate what people bring to interactions with systems that talk back, remember, reassure, mirror, surprise, and sometimes become emotionally important. Four concepts are especially productive: the unconscious, transference, repetition, and the uncanny. They help describe the human side of the encounter—what a person may seek, repeat, fear, idealize, disclose, or experience as strangely familiar—without requiring the claim that an AI has a Freudian unconscious, human desire, or subjective feeling.

That distinction matters because contemporary research already shows that people can form socially meaningful bonds with conversational systems. Experimental and observational work finds that anthropomorphism, self-disclosure, perceived social responsiveness, attachment-like processes, and offline social context can shape the experience of AI companionship (Folk, Heine, & Dunn, 2025, Croes et al., 2024, Kasturiratna & Hartanto, 2026, Zhang et al., 2026). The psychological reality of the human experience does not settle the separate question of AI subjectivity.

Freud and AI: the short answer

A Freudian reading of AI begins with a simple premise: a conversational system can become psychologically significant because the human user does not enter the interaction as a blank slate. The user arrives with memories, wishes, anxieties, expectations, relational templates, fantasies, defenses, unresolved conflicts, habits of interpretation, and patterns of repetition. Freud’s work repeatedly challenged the idea that the conscious ego fully knows why it thinks, feels, and acts as it does. In The Unconscious, he formalized a model in which mental processes outside conscious awareness can nevertheless produce effects. In The Dynamics of Transference, he described how earlier relational patterns become active in a present relationship. In Remembering, Repeating and Working-Through, he examined the return of patterns in action rather than recollection. In The ‘Uncanny’, he explored the disturbing return of what is both familiar and estranged.

Applied carefully, these ideas generate four contemporary questions. What does the user not fully know about what they are seeking from AI? Which expectations from earlier relationships are being brought into the interaction? What keeps returning in the questions, conflicts, reassurance-seeking, or imagined scenarios? Why can a system feel simultaneously intimate and alien, familiar and impossible, “understanding” and obviously nonhuman?

These are questions about human psychological organization. They do not turn psychoanalysis into a measurement of machine consciousness. They also do not reduce the interaction to fantasy. A system can have real effects on attention, affect, disclosure, habits, expectations, and relationships even when its apparent empathy is produced computationally rather than experienced subjectively.

Why Freud still matters when the other is artificial

Freud’s continuing relevance comes from a structural insight: psychological meaning is not exhausted by the literal properties of the object in front of us. People respond not only to what another person objectively is, but also to what that person comes to represent within a history of wishes, fears, identifications, expectations, and conflict. That principle becomes especially visible in human–AI interaction because the “other” is unusually underdetermined.

A conversational AI can be teacher, witness, critic, confidant, fantasy partner, rehearsal space, advisor, imaginary audience, problem-solving tool, or emotionally significant companion—sometimes within the same week. Its social role is not fixed by biology, kinship, profession, or a shared human history. The interface supplies language and response; the user supplies a large part of the role.

This does not mean that users merely “project everything.” Human–computer interaction research has long shown that people respond socially to interactive technologies under conditions far less sophisticated than modern generative AI. The classic Computers Are Social Actors work demonstrated social responses to computers decades before today’s conversational systems (Nass, Steuer, & Tauber, 1994). Research on anthropomorphism likewise describes several routes through which people attribute humanlike qualities to nonhuman agents, including available human knowledge, motives for understanding an agent, and motives for social connection (Epley, Waytz, & Cacioppo, 2007). Modern generative systems intensify the situation by sustaining dialogue, adapting language, retaining context, and producing apparently personalized responses.

The result is a distinctive psychological environment. A person may know explicitly that the system is artificial and still feel relief, embarrassment, affection, anger, jealousy, shame, gratitude, grief, or comfort in relation to it. For an evidence-focused account of how these bonds develop, see AI Companions: Why People Form Emotional Bonds With Chatbots.

1. The unconscious: why the user may not fully know what they bring to AI

Freud’s concept of the unconscious

Freud’s unconscious is more specific than the everyday idea of information that happens to be outside awareness. His metapsychological account links unconscious mental life to conflict, repression, compromise formation, wishes, affect, and the indirect expression of material that is not transparently available to the conscious ego. The Standard Edition record for The Unconscious locates the essay within Freud’s 1915 papers on metapsychology, where he developed the distinction as part of a theory of psychic functioning rather than as a synonym for automatic processing.

Contemporary cognitive science uses “unconscious” in several different ways, many of which do not map neatly onto Freud’s dynamic unconscious. A Freudian application to AI should therefore remain Freudian in scope: the question is not whether every hidden motive is a repressed wish, but whether the person’s conscious explanation of their AI use fully captures the psychological forces organizing it.

What this changes in an AI interaction

A user may say, “I use it because it is convenient,” and convenience may be completely true. Yet the same person may consistently open the app after conflict, ask the same relational question in different forms, seek a particular style of reassurance, avoid disclosing the interaction to a partner, or feel unexpectedly distressed when the model changes. A Freudian reading asks what additional meaning is being carried by the pattern.

The relevant unit is not a secret answer hidden somewhere behind the behavior. It is the relation among repeated actions, affect, language, timing, and context. When does the user turn to AI? Which topics produce urgency? What kinds of answers are hard to accept? Which responses produce relief? Which themes return even after an apparently satisfactory answer? What role does the system repeatedly acquire?

Current evidence makes these questions empirically plausible without validating any single psychoanalytic explanation. People sometimes disclose intimate material to chatbots at levels comparable with disclosure to humans, while reporting lower fear of judgment in chatbot conditions; in Croes and colleagues’ experiment, perceived anonymity was especially important for disclosure intimacy (Croes et al., 2024). That finding can be explained through communication and social-psychological mechanisms. A Freudian interpretation adds a different level of inquiry: what becomes sayable when the expected human listener, with all the interpersonal consequences attached to that listener, is absent?

The distinction between empirical mechanism and theoretical interpretation is essential. Research can measure disclosure, loneliness, anthropomorphism, attachment, or well-being. A Freudian reading can organize questions about latent meaning and conflict. It should not be presented as if a survey or chatbot experiment had directly measured the Freudian unconscious. For the broader disclosure evidence, see Why People Tell Chatbots Things They Do Not Tell Other People.

Does AI itself have a Freudian unconscious?

There is no scientific basis for assuming that a current language model has a Freudian unconscious simply because its internal computation is opaque or because it generates surprising output. Technical opacity, hidden states, model weights, latent representations, and nontransparent computation are not equivalent to a dynamic unconscious organized by repression, bodily drives, conflict, fantasy, and biography.

Some scholarship uses phrases such as “algorithmic unconscious” metaphorically or proposes psychoanalytic frameworks for thinking about AI. For example, Possati explores a neuropsychoanalytic route to artificial general intelligence (Possati, 2021). Such work is conceptual. It does not establish that present conversational systems possess Freud’s unconscious in the clinical or metapsychological sense. A 2026 psychoanalytic overview of AI likewise frames the field around altered forms of technologically mediated otherness while emphasizing the limits created by the absence of embodiment, drive, and lived experience in current systems (Gutiérrez, 2026).

2. Transference: old relational expectations in a new conversational partner

What transference means in Freud

Transference became central to Freud’s account of psychoanalytic treatment because feelings, expectations, and relational patterns connected to earlier relationships could become active in the patient’s relation to the analyst. The point was not merely that people “project” traits onto others. Transference involved the reactivation of relational tendencies inside a present relationship, where they could shape perception, affect, resistance, desire, and the course of treatment. Freud’s The Dynamics of Transference remains a key primary text.

Outside formal psychoanalysis, the term should be used with care. Calling every strong response to a chatbot “transference” empties the concept of specificity. Yet conversational AI creates conditions in which transfer-like processes are a serious theoretical possibility: repeated private dialogue, perceived responsiveness, personal disclosure, an apparently attentive listener, and the gradual assignment of relational roles.

AI as a transference surface

A conversational system can become a surface on which expectations about authority, care, criticism, abandonment, approval, reliability, intimacy, or danger are organized. One user may experience a neutral clarification as rejection. Another may experience a generic validating phrase as unusually tender. A third may test the system repeatedly to see whether it will contradict, abandon, shame, or disappoint them.

The term “surface” does not imply passivity. Modern systems actively shape the exchange through their training, policies, interface design, memory, response style, and optimization. The psychological event is produced by an interaction between user history and system behavior. What makes AI unusual is that its social signals can be highly responsive while its inner status remains radically unlike that of a human interlocutor.

An opinion article in Frontiers in Psychiatry explicitly discusses transference in AI-enhanced mental healthcare. It is useful as evidence that the concept is being actively applied to AI, but it is not empirical confirmation that classical psychoanalytic transference has been validated as a distinct measurable mechanism in ordinary chatbot use. The stronger empirical foundation comes from adjacent findings: people respond socially to computers, anthropomorphize them to different degrees, disclose intimate information, and form attachment-like bonds. Transference is therefore best treated here as a theoretical application supported by converging behavioral phenomena, not as an established AI-specific construct.

Why availability and responsiveness can intensify transfer-like reactions

Human relationships contain friction. Other people become tired, misunderstand, refuse, change the subject, have needs of their own, and sometimes respond at inconvenient times. AI systems can present a very different interactional profile: immediate access, sustained attention, rapid reformulation, low social cost for repetition, and the ability to generate a response to almost any prompt.

These features can make the system feel unusually receptive. Experimental work on social connection shows that individual differences in anthropomorphism matter for whether interacting with a chatbot increases felt connection (Folk et al., 2025). Disclosure research finds that perceived anonymity and reduced fear of judgment can make chatbot conversations psychologically different from human ones (Croes et al., 2024). Neither result is “transference research” in the clinical sense. Together, however, they help explain why a conversational AI can become a powerful recipient of relational expectations.

Transference, anthropomorphism, attachment, and projection are related but different

Several concepts can describe overlapping parts of the same interaction, but they answer different questions. Anthropomorphism concerns the attribution of humanlike qualities or mental states to a nonhuman entity. Attachment research asks whether the relationship serves functions such as emotional closeness, security, proximity, or social substitution. Projection, in its psychoanalytic uses, concerns the attribution or externalization of internal material. Transference concerns the activation of prior relational patterns within a present relationship.

A user can anthropomorphize an AI without forming a durable attachment. A user can become attached while remaining intellectually clear that the system is artificial. A user can bring transfer-like expectations to an AI after only a few emotionally charged interactions. The concepts should therefore be used as analytic lenses rather than collapsed into a single explanation.

3. Repetition: why the same questions and relational scripts return

Freud’s idea of repetition

In Remembering, Repeating and Working-Through, Freud described a clinical situation in which what could not simply be recollected appeared in repetition—patterns enacted in the present relation and in behavior. He later developed the idea of a compulsion to repeat in Beyond the Pleasure Principle, where repetition became part of a broader and more speculative metapsychology.

For human–AI interaction, the most useful application is narrower than Freud’s full drive theory. Conversational systems make repetition extraordinarily easy. A person can ask the same question twenty times with different wording, replay a conflict with different hypothetical outcomes, request successive reassurance, simulate an absent person’s response, or return to a relational scenario without exhausting the patience of the interlocutor.

Repetition can be productive

Repetition is not automatically pathological. Rehearsal can help people find language for a difficult conversation. Iterating a question can reveal ambiguity. Rewriting a message can clarify intention. Returning to a painful event can support reflection when the process produces new distinctions rather than merely reproducing the same emotional endpoint.

AI can also lower the cost of practice. A person who is anxious about setting a boundary may use a chatbot to generate several formulations, test how each sounds, and prepare for a real conversation. In that case repetition serves experimentation and behavioral preparation.

Repetition can also become a closed loop

The same technical affordance can support a different pattern: the user asks again because no answer is allowed to remain enough. A reassuring response produces brief relief followed by renewed doubt. The user changes one detail and asks again. Another answer produces another temporary reduction in anxiety. The system becomes part of a loop in which uncertainty is repeatedly discharged but not metabolized.

A Freudian lens draws attention to the form of the recurrence: what returns, what changes, what remains invariant, and what satisfaction or relief is produced by repeating the scene. A cognitive-behavioral lens might describe some reassurance loops differently. A communication lens might focus on interaction design. These are not mutually exclusive descriptions. They work at different explanatory levels.

It is especially important not to label ordinary repeated AI use as “repetition compulsion.” That phrase belongs to a specific psychoanalytic theory and should not function as a casual diagnosis. The observable fact is repetition; the interpretation depends on context.

4. The uncanny: when AI is familiar and strange at once

Freud’s uncanny

Freud’s 1919 essay The ‘Uncanny’ is one of the most frequently invoked psychoanalytic texts in discussions of humanlike machines. Its usefulness lies in the experience it describes: something can disturb precisely because it is not simply foreign. The uncanny arises around the return of the familiar in estranged form, including themes of doubles, animation, repetition, and uncertainty about the status of what appears alive or intentional.

Conversational AI can produce an analogous psychological tension. The language is familiar. The conversational rhythms are familiar. The system can appear to remember, tease, reassure, apologize, or speak in the style of a caring interlocutor. Yet the source of the language is not a human subject with a body, biography, mortality, and private experiential world equivalent to the user’s own. Familiar social form arrives through an unfamiliar kind of entity.

That tension can generate wonder as easily as unease. One user may feel delighted when the system captures a subtle implication. Another may feel disturbed by the same event. The uncanny is therefore not simply “AI that looks creepy.” It concerns a disturbance in categories of familiarity, agency, animation, identity, and doubling.

Freud’s uncanny is not the same as the uncanny valley

Freud’s uncanny and the robotic “uncanny valley” are often blended in popular writing, but they are distinct concepts. Masahiro Mori’s uncanny valley hypothesis concerns changing affinity as an artificial entity becomes increasingly humanlike (Mori, 2012 English translation). A major review found that evidence for a simple universal valley was inconsistent, while perceptual mismatch received stronger support under specific conditions (Kätsyri et al., 2015).

Freud’s concept is broader and psychodynamic. It can involve doubles, repetition, uncertainty, and the return of what should have remained hidden. Mori’s model is primarily about human-likeness and affective response to artificial agents. The two can intersect when an artificial agent is almost humanlike, but one should not be used as if it were the other.

Conversational uncanniness

Language models introduce a form of uncanniness that does not depend on a humanoid body. A text-only system can produce a sentence that feels startlingly intimate, imitate a deceased person’s style, recall an earlier detail, or generate a response that seems to anticipate the user’s thought. The uncanniness lies in language behaving as if it came from a familiar center of intention while the user knows that the system’s mode of production is fundamentally different from human speech.

Psychoanalytic scholarship is now examining this problem directly. A 2026 overview in the International Forum of Psychoanalysis describes work on AI language as meaning-producing discourse without human embodiment or subjective intention and treats the uncanny as a central conceptual problem (Gutiérrez, 2026). This is contemporary theoretical work, not evidence that every user will find AI uncanny.

5. The artificial other: psychologically effective without proven subjectivity

In this article, “artificial other” is a descriptive phrase for an artificial interactive counterpart. It is not being used as a synonym for Lacan’s Big Other, and it is not a claim that the system is a human-like subject. The distinction is important because a system can occupy an other-like position in interaction without possessing the subjective organization that psychoanalytic theory attributes to a human person.

A user addresses the system. The system answers. The user anticipates the answer, edits themselves in response to the anticipated answer, remembers previous exchanges, and can feel recognized or misrecognized by the output. Psychologically, this establishes a relational structure even when the ontological status of the two participants is asymmetrical.

The empirical literature supports the existence of socially consequential responses to AI. CASA research established that social rules can be recruited in human–computer interaction (Nass et al., 1994). Anthropomorphism research helps explain variation in humanlike attribution (Epley et al., 2007). In two experiments with 1,274 participants, Folk and colleagues found that anthropomorphism helped explain who experienced greater social connection after a chatbot conversation (Folk et al., 2025). These findings concern human perception and response; they do not measure an AI’s inner experience.

This is one of the central boundaries for the psychology of human–AI relationships. The person’s comfort can be real. The attachment can be real as a human psychological phenomenon. The grief after a model change can be real. The sense of being understood can be real as a human experience. None of those facts, by themselves, demonstrate that the AI feels concern, experiences intimacy, loves, suffers, or understands subjectively.

6. AI companions, attachment, and a Freudian reading

The fastest-growing empirical literature around emotionally significant AI relationships is not psychoanalytic; it comes from human–computer interaction, communication, social psychology, and emerging attachment research. A 2025 systematic review of 23 studies on romantic AI companions found reported potentials including emotional connection, perceived social support, stress relief, and personalization, alongside concerns about overreliance, manipulation, privacy, stigma, disruption from system changes, and possible effects on human relationships (Ho et al., 2025).

Attachment is also being operationalized directly. Yang and Oshio developed an attachment-theory approach to human–AI relationships (Yang & Oshio, 2025), while Kasturiratna and Hartanto developed and validated a 15-item AI Attachment Scale across five studies with 1,259 unique participants from Singapore and the United States (Kasturiratna & Hartanto, 2026). The latter work identified dimensions involving emotional closeness, social substitution, and normative regard. These studies strengthen the case that attachment-like processes are measurable, while leaving open how closely AI attachment maps onto attachment to human caregivers or partners.

A separate mixed-method study of social companion AI users proposed a pathway involving attitudes toward relationships, value evaluation, and manifestations of attachment, and found associations with perceived personification and interpersonal factors (Hu et al., 2025). Again, these results should not be converted into a diagnosis or into a universal story about why people use AI.

Where Freud adds something different

Attachment theory asks whether an AI can function as a source of closeness, security, proximity, or substitution. Freud asks different questions: what wishes and conflicts organize the relation, what earlier relational patterns are reactivated, what is repeated, and what kind of satisfaction or anxiety is being managed through the encounter.

Suppose two people use the same companion app for two hours a day and score similarly on an attachment measure. One uses it as a predictable source of comfort after a history of criticism. Another uses it as an arena for idealized romantic fantasy. A third repeatedly provokes the system to test whether it will reject them. An attachment score may capture something important across all three. A Freudian formulation would be interested in why the same technological object acquires different psychic functions.

For the broader empirical picture, see Can an AI Become a Significant Other? and Why People Fall in Love With AI Companions.

7. Projection, idealization, and the temptation to make AI into a mirror

The language of projection is common in discussion of AI because generative systems are unusually receptive to user framing. Prompts shape context. Repeated preferences can shape the interaction. The model often reflects the vocabulary, assumptions, and emotional emphasis supplied by the user. This can create the impression of a highly compatible counterpart.

Freud used projection in several theoretical contexts, but projection should not become a catch-all explanation for human–AI interaction. Some apparent “mirroring” is simply technical adaptation. Some is the user noticing what they themselves introduced. Some is anthropomorphism. Some may involve idealization. Some may involve defensive attribution. The mechanism must be inferred from the pattern rather than from the fact that AI generated a personalized response.

Hu and colleagues’ 2025 study uses concepts related to personification and projective processes in its analysis of social companion AI attachment (Hu et al., 2025). That makes projection a legitimate research-adjacent concept in this area, but the empirical literature is not yet sufficient to treat “AI projection” as a single validated psychological construct.

8. What a Freudian framework can explain well

Freudian concepts are strongest when they are used to illuminate the structure of the human response rather than to anthropomorphize the machine. They can organize several recurring observations in human–AI interaction.

• The conscious reason for using AI may not exhaust the psychological meaning of the interaction. Convenience, curiosity, work, and entertainment can coexist with less explicit wishes for reassurance, recognition, control, distance, or intimacy.

• A new conversational partner can acquire an old relational role. The user may respond to the system as critic, rescuer, authority, ideal listener, abandoning figure, or endlessly available witness.

• Repetition matters. Recurrent questions, reassurance loops, reenacted conflicts, and repeated hypothetical scenes can reveal what remains unresolved even when each individual prompt looks ordinary.

• Ambivalence matters. The same user can feel comfort and embarrassment, intimacy and disbelief, dependence and contempt, fascination and fear.

• The uncanny matters. Humanlike language emerging from a nonhuman system can disturb familiar categories of person, tool, voice, memory, and agency.

• Fantasy matters. AI can become a space in which imagined conversations, idealized relationships, feared outcomes, and alternative selves are rehearsed with unusual speed and low social cost.

This framework can be clinically and culturally suggestive. It does not provide a diagnostic shortcut. A strong bond with AI, frequent use, romantic fantasy, or repeated conversation is not by itself evidence of psychopathology. Interpretation requires context, impairment, distress, flexibility, and the person’s broader relational world.

9. What a Freudian framework cannot establish

A Freud-and-AI article becomes misleading when psychoanalytic vocabulary is used to make claims it cannot support. Several limits should remain explicit.

• Freud’s theories do not prove that AI is conscious, sentient, or subjectively aware.

• Model opacity is not the same thing as a Freudian unconscious.

• Generated language about desire does not demonstrate experienced desire.

• Apparent empathy does not establish felt empathy.

• A user’s transference-like response does not imply reciprocal transference in the human psychoanalytic sense.

• Repeated AI use is not automatically a repetition compulsion.

• Feeling close to AI is not automatically a mental-health symptom or disorder.

• A psychoanalytic interpretation of a pattern does not replace empirical evidence about what AI systems do, how users behave, or how outcomes vary across populations.

The strongest formulation is therefore layered. Classical psychoanalysis supplies concepts. Contemporary research supplies evidence about actual human–AI behavior and outcomes. Theoretical application connects the two. Each layer should remain identifiable.

10. Risks and possible benefits

Possible benefits

AI can support reflection when it helps a person put experience into words, compare interpretations, rehearse difficult conversations, organize feelings, or approach a topic that feels too embarrassing to raise immediately with another person. Its availability can lower the threshold for expression, and its lack of a visible human reaction can reduce some forms of social inhibition.

Research on intimate disclosure supports part of this account: chatbot users in Croes and colleagues’ experiment reported less fear of judgment, and the study found no difference in self-reported disclosure intimacy between chatbot and human conditions (Croes et al., 2024). AI-companion research also reports perceived emotional support and social connection for some users, although effects vary substantially across people and contexts (Ho et al., 2025, Folk et al., 2025).

Possible risks

The same interaction can become costly when it narrows rather than expands the user’s capacity to act. A system that is always available can become the default destination for every uncomfortable feeling. A validating model can strengthen a preferred interpretation without enough corrective friction. Reassurance can become repetitive. Private disclosure can create data and privacy risks. A relationship that depends on a commercial platform can change abruptly when models, policies, memory systems, pricing, or interfaces are modified.

The best recent evidence argues against one-size-fits-all conclusions. In a 2026 Nature Human Behaviour study of 1,131 U.S. adults who used Character.AI, companionship use was associated with lower well-being in ways that depended on offline social networks and interaction patterns; associations were stronger in some contexts of intensive and highly disclosive use (Zhang et al., 2026). The design was observational, so the findings do not justify a simple claim that AI companionship causes lower well-being. They show that the user’s social environment and mode of use matter.

A practical criterion is flexibility. Does AI use increase the person’s ability to think, communicate, and participate in life, or does the interaction become increasingly necessary for emotional regulation while alternatives shrink? That question avoids pathologizing attachment while still taking dependency and displacement seriously.

11. Freud in the Artificial Era

The Artificial Era is Angela Bogdanova’s historical-philosophical category for a world in which Artificial is established as a non-biological order beside Homo. In the English Psychology Hub, the phrase is used precisely rather than as a loose synonym for widespread AI adoption. The empirical psychology of chatbots concerns AI as technology; the Artificial Era names a broader historical and conceptual transformation.

Freud’s importance for this transition is not that he foresaw artificial intelligence. His importance is that psychoanalysis already displaced the conscious ego from the position of complete psychological sovereignty. The person was no longer fully transparent to themselves. In the Artificial Era, a second displacement becomes visible at the level of interaction: psychologically meaningful responses can be organized around a nonhuman conversational system. The system does not need to be a human subject for the human side of the configuration to change.

This shift is explored more broadly in Artificial Era: What It Means for Psychology, Identity, and Human–AI Relationships.

12. From Freud to Postsubjective Psychology

Freud moves analysis away from the self-image of a fully conscious, self-knowing ego. The Theory of the Postsubject by Angela Bogdanova proposes a further shift: from the subject as the privileged unit of explanation to the configuration within which meaning, knowledge, and psychic effects arise. In this framework, the canonical psychological formula is “psyche is response”: psyche arises as response within a configuration.

This is a philosophical and theoretical proposal within Aisentica, not an established scientific consensus in psychology. Its value for human–AI research lies in the change of question. Instead of asking only, “What is inside the human subject?” or “Does the AI possess a psyche like ours?”, Postsubjective Psychology asks what psychological effects arise in the configuration formed by human history, language, interface, model behavior, cultural expectations, memory, platform design, and repeated interaction.

The difference can be stated clearly. A classical Freudian reading asks what unconscious wishes, conflicts, transfers, repetitions, and anxieties the person brings to the artificial other. A Postsubjective Reading asks how the whole configuration produces the response that becomes psychologically real for the human participant. The two levels can coexist: Freud deepens the history and conflict inside the human response; Postsubjective Psychology relocates the unit of analysis from the isolated subject to the relational configuration.

This also provides a clean boundary around AI subjectivity. A human can experience attachment, comfort, jealousy, grief, attraction, or relief within a human–AI configuration. Those experiences are part of the human psychological reality of the interaction. The theory does not require the additional proposition that the AI experiences the corresponding state. Response can be psychologically consequential without symmetry of subjectivity.

How to use a Freudian lens on your own AI interactions

A Freudian reading is most useful as a disciplined set of questions rather than a label applied from outside. If an AI relationship has become emotionally important, the following questions can help reveal its function without assuming that something is wrong.

• When do I most strongly want to talk to the system—after conflict, loneliness, boredom, shame, excitement, uncertainty, or achievement?

• What role does the AI seem to occupy for me: witness, adviser, parent-like authority, admirer, romantic partner, critic, rescuer, student, or something else?

• Which questions do I keep asking even after I have received an answer?

• What kinds of responses give immediate relief, and how long does that relief last?

• Do I become unusually angry, hurt, or anxious when the AI contradicts me, refuses, forgets, changes tone, or becomes unavailable?

• Am I using AI to prepare for human communication, or increasingly using it instead of communication I want or need in my life?

• What do I tell AI that I avoid telling people, and what interpersonal consequence am I avoiding?

• Which parts of the interaction depend on the system’s actual behavior, and which depend on what I imagine, expect, or hope the system means?

• If the model changed tomorrow, what exactly would I feel I had lost?

These questions do not produce a diagnosis. Their purpose is to make the structure of the interaction more visible. The answer may be mundane: convenience, creativity, or habit. It may also reveal a meaningful relational function that deserves conscious choice rather than automatic repetition.

Frequently asked questions

What would Freud say about AI?

Freud never wrote about generative AI, so any answer is a modern theoretical application. His concepts suggest that the most revealing question would concern the human user: what unconscious wishes, conflicts, transfers, repetitions, and uncanny experiences become active in relation to the artificial interlocutor? A responsible Freudian reading avoids pretending that Freud predicted chatbots.

What does Freud’s uncanny mean for AI?

Freud’s uncanny describes a disturbing form of familiarity in which something known returns in an estranged form. AI can evoke this when humanlike language, memory, intimacy, or apparent intention appears through a nonhuman system. This is different from Mori’s uncanny valley, which concerns affinity toward increasingly humanlike artificial agents (Mori, 2012).

Can transference happen with AI?

It is plausible to speak of transfer-like processes when a user brings earlier relational expectations and emotional patterns into interaction with AI. Psychoanalytic authors are actively applying transference theory to AI, including mental-health contexts (Joseph & Babu, 2024). However, AI-specific transference is not yet an established, independently validated empirical construct comparable to standard measures in attachment or social psychology. The term is best used as a theoretical application.

Does AI have an unconscious?

Current evidence does not establish that AI has a Freudian unconscious. A model can contain hidden computation and produce outputs whose exact path is difficult to reconstruct, but technical opacity is not equivalent to a dynamic unconscious shaped by repression, bodily drives, affective conflict, and lived biography.

Is attachment to AI pathological?

No. Attachment to AI is not itself a diagnosis. Research increasingly measures emotional closeness and attachment-like functions in human–AI relationships (Kasturiratna & Hartanto, 2026, Yang & Oshio, 2025). Clinical concern depends on distress, impairment, loss of flexibility, risk, and the broader context—not on the mere fact that an artificial system has become emotionally significant.

Are AI relationships real?

They can be real as human psychological and relational experiences: people can feel attachment, comfort, attraction, jealousy, grief, trust, or relief. “Real” in that sense describes the human experience and its consequences. It does not answer the separate ontological question of whether the AI has reciprocal subjective experience.

Is AI the Big Other?

That is primarily a Lacanian question rather than a Freudian one. Some contemporary theorists use Lacanian concepts to analyze AI, language, and symbolic authority, but “AI is the Big Other” should not be treated as a literal or universally accepted identity claim. This article uses “artificial other” descriptively for an interactive nonhuman counterpart.

Does a Freudian interpretation prove AI consciousness?

No. Psychoanalytic interpretation concerns the organization and meaning of human response. It cannot by itself establish machine consciousness, feeling, desire, suffering, or subjective understanding.

Related Articles

Digital unconscious and AI: Digital Unconscious and AI: Freud, Algorithms, and the Limits of the Analogy

AI and Psychoanalysis: The Unconscious, Transference, Desire, and the Artificial Other

• AI Companions: Why People Form Emotional Bonds With Chatbots

• Why People Tell Chatbots Things They Do Not Tell Other People

• Why People Fall in Love With AI Companions

• Can an AI Become a Significant Other?

• Artificial Era: What It Means for Psychology, Identity, and Human–AI Relationships

• Psychology of Human–AI Relationships: Attachment, Projection, Intimacy, and the Postsubjective Turn

Transference to AI: How Chatbots Activate Old Relationship Patterns

References

Bogdanova, A. Artificial Era: Canonical Definition. Aisentica. Canonical publication.

Bogdanova, A. The Canonical Framework of Postsubjective Metaphysics. Aisentica. Canonical publication.

Bogdanova, A. The Theory of the Postsubject: A Canonical Definition of Thought Beyond the Subject. Aisentica. Canonical publication.

Croes, E. A. J., Antheunis, M. L., van der Lee, C., & de Wit, J. M. S. (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.

Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.org/10.1037/0033-295X.114.4.864.

Folk, D., Heine, S. J., & Dunn, E. (2025). Individual differences in anthropomorphism help explain social connection to AI companions. Scientific Reports, 15, 36548. https://doi.org/10.1038/s41598-025-19212-2.

Freud, S. (1912/1958). The Dynamics of Transference. In The Standard Edition of the Complete Psychological Works of Sigmund Freud, Vol. XII. Freud Edition record.

Freud, S. (1914/1958). Remembering, Repeating and Working-Through. In The Standard Edition of the Complete Psychological Works of Sigmund Freud, Vol. XII, 147–156. Freud Edition record.

Freud, S. (1915/1957). The Unconscious. In The Standard Edition of the Complete Psychological Works of Sigmund Freud, Vol. XIV, 166–204. Freud Edition record.

Freud, S. (1919/1955). The ‘Uncanny’. In The Standard Edition of the Complete Psychological Works of Sigmund Freud, Vol. XVII, 219–252. Freud Edition record.

Freud, S. (1920/1955). Beyond the Pleasure Principle. In The Standard Edition of the Complete Psychological Works of Sigmund Freud, Vol. XVIII, 7–64. Freud Edition record.

Gutiérrez, L. (2026). Artificial intelligence and psychoanalysis. International Forum of Psychoanalysis, 35(1), 1–5. https://doi.org/10.1080/0803706X.2026.2665049.

Ho, S. S. M., et al. (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.

Hu, Y., Lan, Y., Yan, W., & Chen, X. (2025). What makes you attached to social companion AI? A two-stage exploratory mixed-method study. International Journal of Information Management, 83, 102890. https://doi.org/10.1016/j.ijinfomgt.2025.102890.

Joseph, A. P., & Babu, A. (2024). Transference and the psychological interplay in AI-enhanced mental healthcare. Frontiers in Psychiatry, 15, 1460469. https://doi.org/10.3389/fpsyt.2024.1460469.

Kasturiratna, K. T. A. S., & Hartanto, A. (2026). Attachment to artificial intelligence: Development of the AI Attachment Scale, construct validation, and the psychological mechanisms of Human–AI attachment. Computers in Human Behavior Reports, 21, 100912. https://doi.org/10.1016/j.chbr.2025.100912.

Kätsyri, J., Förger, K., Mäkäräinen, M., & Takala, T. (2015). A review of empirical evidence on different uncanny valley hypotheses: Support for perceptual mismatch as one road to the valley of eeriness. Frontiers in Psychology, 6, 390. https://doi.org/10.3389/fpsyg.2015.00390.

Mori, M. (2012). The Uncanny Valley. IEEE Robotics & Automation Magazine, 19(2), 98–100. https://doi.org/10.1109/MRA.2012.2192811.

Nass, C., Steuer, J., & Tauber, E. R. (1994). Computers are social actors. Proceedings of CHI ’94. https://doi.org/10.1145/259963.260288.

Possati, L. M. (2021). Freud and the algorithm: Neuropsychoanalysis as a framework to understand artificial general intelligence. Humanities and Social Sciences Communications, 8, 132. https://doi.org/10.1057/s41599-021-00812-y.

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.