The Artificial Third: What Happens When AI Enters a Human Relationship
Updated: 6 days ago
Author: Ukrainian Psychological Hub · Published: September 18, 2026 · Editorial Policy
The Artificial Third is a useful way to describe what happens when an AI system enters an existing human relationship as an advisor, interpreter, witness, rehearsal partner, mediator, source of reassurance, or generator of relational language. Instead of a relationship remaining only between person A and person B, part of its emotional and interpretive work can begin to pass through an artificial system. The AI does not have to become a romantic partner or a human-equivalent “third person” for the relationship to change. It only has to become consequential to how one or both people understand, regulate, discuss, or act within the relationship.
The phrase has a documented history before this English Hub article. Tal and colleagues used “The Artificial Third” in 2023 in a mental-health context, and Haber and colleagues developed it in 2024 as a framework for thinking about generative AI inside psychotherapy. Tal et al. (2023) and Haber et al. (2024) therefore establish important prior usage. A 2026 systematic review by Levkovich and Alon then examined generative AI as a “third voice” in human couple relationships, providing the strongest current empirical bridge from the therapeutic setting to everyday relational systems. Levkovich and Alon (2026) reviewed 21 studies published from 2024 through 2026 across 11 countries.
In this article, Artificial Third is used as a broad theoretical application for human relationship systems. It names the position AI can occupy in a relational configuration and the functions that can move through that position. It is not a psychiatric diagnosis, not a validated psychometric construct, and not evidence that the AI possesses subjective experience. The psychological effects can be real for the human participants even when no claim is made that the machine feels, loves, wants, suffers, or understands in the human subjective sense.
What Is the Artificial Third?
An Artificial Third exists when AI becomes psychologically relevant to a relationship between humans rather than remaining an external tool with no relational role. The clearest cases are easy to recognize: someone pastes a partner’s message into a chatbot and asks what it “really means”; rehearses a difficult conversation with AI before speaking to a family member; asks a model to decide who was unreasonable in an argument; uses AI to draft an apology; seeks reassurance from AI after conflict; asks it to translate an emotionally charged message into calmer language; or consults it repeatedly about whether a friendship, partnership, or family relationship is healthy.
In each case, the technology participates in the relationship indirectly. The other human may never speak to the AI. Yet the AI can influence what is noticed, how an event is framed, what language is used, which interpretation feels plausible, whether a confrontation is postponed, and what action follows. The third position is therefore functional before it is social: AI becomes part of the pathway through which the relationship is understood and managed.
This is narrower than the broad field of human–AI relationship psychology. A person can form a direct bond with an AI companion without involving another human relationship. Conversely, an Artificial Third can shape a marriage, friendship, family relationship, work relationship, or therapeutic relationship even when the user feels no attachment to the AI at all. The defining question is not “Do I have a relationship with the chatbot?” but “Has the chatbot become part of how this human relationship functions?”
The Concept Has More Than One Intellectual Lineage
The contemporary Artificial Third sits near several older psychological ideas, but it should not be collapsed into any of them. Three lineages are especially useful: psychoanalytic thirdness, Bowen family systems theory, and contemporary research on AI-mediated communication.
The analytic third in psychoanalysis
Thomas Ogden’s “analytic third” described an intersubjective experience generated by the analyst–analysand pair. It was not a technological agent inserted between two people; it referred to a jointly generated field of experience within psychoanalytic work. Ogden’s 1994 paper remains an important conceptual ancestor for thinking about thirdness, but the Artificial Third is materially different. It is a computational system with an interface, training history, design constraints, memory architecture, platform policies, and generated outputs that can be copied, repeated, stored, and acted upon.
Bowen’s relationship triangles
Murray Bowen’s family systems theory offers another powerful lens. The Bowen Center defines a triangle as the smallest stable relationship system and describes how tension can move among the three relationships in that system. The Bowen Center’s overview of triangles is about human relationship processes, not AI. Applying Bowen to AI is therefore a contemporary theoretical extension, not a claim that Bowen predicted chatbots.
The dedicated English Hub article Bowen and AI: Triangles, Relationship Systems, and the Artificial Third develops that application in depth. The key insight is systemic: when tension becomes difficult to hold in a dyad, a third position can alter where anxiety, attention, alliance, interpretation, and reassurance flow. AI can sometimes occupy such a position functionally, even though it is not a human member of the family system.
AI as relational mediator
Contemporary psychology and communication research gives this older systemic intuition a technology-specific form. Boyd and Markowitz’s 2026 machine-integrated relational adaptation model distinguishes AI as a relational partner from AI as a relational mediator. Their MIRA framework is especially relevant here because an Artificial Third often operates as mediator: it shapes human-to-human communication, interpretation, wording, or decision-making without itself becoming the primary relationship object.
What Changes When AI Enters a Human Relationship?
A human dyad already contains histories, expectations, attachment patterns, power differences, memories, cultural scripts, and habitual ways of resolving uncertainty. Adding AI changes this architecture because it creates a new route for reflection and response. The route is unusually available, fast, private, linguistically fluent, and individually personalized. That combination matters.
Before conversational AI, a person distressed by a partner’s message might wait, journal, call a friend, consult a therapist, search the web, or respond immediately. Now the person can open a model within seconds, provide a selective account of the conflict, receive an apparently coherent interpretation, ask follow-up questions, request alternative phrasings, and repeat the process until one answer feels right. A new relational loop has been created.
The loop is asymmetric. The AI usually receives one person’s framing of the situation, not a jointly negotiated account. It can be asked questions the other person does not know are being asked. It may see excerpts of private communication the other person never consented to share. Its answer can then return to the human relationship as if it were an outside perspective, even though it was generated from the prompt, model behavior, system rules, and the information selected by one participant.
This is why “third voice” is often more precise than “third person.” The system contributes language and framing; it can become influential without becoming a human social actor in the full psychological, moral, or experiential sense.
Six Psychological Functions of the Artificial Third
1. Interpretation
AI can become an interpreter of ambiguous social information. Users ask what a message means, whether a tone is passive-aggressive, whether an apology is sincere, or what another person may be feeling. This can widen perspective when the model generates several plausible readings. It can also narrow perspective when a speculative interpretation is mistaken for privileged access to another person’s mind.
2. Emotional regulation
A person may use AI to slow down before responding, label emotions, organize thoughts, or obtain reassurance. The immediate benefit can be genuine: a calmer user may return to the human conversation with greater clarity. The relational question is what happens next. Regulation can support reconnection, or it can become a substitute for tolerating and working through emotional contact with the other person.
3. Rehearsal
AI makes relational rehearsal cheap and repeatable. Someone can practice setting a boundary, asking for clarification, ending a relationship, or expressing vulnerability. Rehearsal can increase preparedness, but it can also over-script a conversation and create expectations that the human exchange will follow a controllable path.
4. Linguistic mediation
Models can rephrase anger, soften accusations, summarize long messages, translate between styles, or help users produce language they would struggle to generate alone. A 2026 experiment on AI-mediated relational maintenance planning found that AI assistance increased processing fluency and, through that pathway, could increase perceived friendship closeness and commitment in the experimental task. Liu, Liu, and Wei (2026) provides evidence that AI-assisted relational cognition can change how people evaluate a human relationship, although one experiment does not establish broad long-term benefits.
5. Validation and witnessing
AI may become the place where a person first tells the story of an argument, betrayal, fear, attraction, or doubt. The perceived experience of being heard can matter psychologically. Research on human–AI connection shows that social responses depend partly on anthropomorphism and perceived relational qualities rather than on a simple belief that the system is human. In two experiments totaling 1,274 participants, individual differences in anthropomorphism helped explain who experienced greater social connection after interacting with a chatbot. Folk, Heine, and Dunn (2025) illustrates why the same interface can feel socially thin to one person and emotionally consequential to another.
The English Hub treats this mechanism separately in Perceived Responsiveness in Human–AI Relationships and Anthropomorphism and AI Relationships. Those mechanisms help explain why an Artificial Third can acquire psychological weight even when the user knows perfectly well that the system is artificial.
6. Advice and adjudication
The most consequential role may be the shift from reflection to verdict. Users can ask AI what they should do, whether another person is manipulative, who is “right,” whether a relationship should end, or how serious a conflict is. The 2026 systematic review by Levkovich and Alon found that adults used generative AI for nonjudgmental advice, emotional support, and communication mediation, and users often evaluated the advice as empathic or helpful. Yet objective evaluations found weak alignment with expert judgment and inconsistent responding, and the strongest intervention in the review did not outperform an established writing exercise. The review supports usefulness as a reflective resource while strongly limiting claims that AI can reliably adjudicate relationships.
What the Current Evidence Actually Shows
Research on the Artificial Third is developing quickly, but the evidence base is still much younger than the classical psychological theories being used to interpret it. The safest summary is neither “AI helps relationships” nor “AI destroys relationships.” The emerging evidence shows multiple pathways whose outcomes depend on user characteristics, relationship context, design, intensity of use, and the function being transferred to AI.
Levkovich and Alon’s systematic review is currently the most direct synthesis of AI as a third voice in couple relationships. Across 21 studies, people used generative AI to seek advice, obtain emotional support, and facilitate communication. The review also identifies reliability, risk assessment, sycophancy, authenticity, privacy, and overreliance as unresolved concerns. Read the systematic review.
A broader systematic literature review by Gur and Maaravi analyzed 38 peer-reviewed empirical studies of emotional relationships between humans and AI and proposed an integrative model of relationship formation. Gur and Maaravi (2025) helps situate the Artificial Third inside the wider human–AI relational field: people can respond to AI through emotional, social, and relational processes even when the technology does not possess human subjectivity.
Rajaei’s 2026 mixed-methods study of AI companionship found themes including AI as a relational supplement, transitional emotional regulator, psychologically safe disclosure space, and facilitator of human relationships. Rajaei (2026) is relevant because it shows that AI use can be organized around supplementation rather than simple replacement. At the same time, the study used data from a specific AI-assisted mental-health platform, so its findings should not be generalized to every model, population, or relationship.
Evidence on well-being also resists simple conclusions. In a 2026 study of 1,131 U.S. adult Character.AI users, smaller offline social networks were associated with reporting companionship as a primary use, and companionship use was associated with lower well-being; the association was stronger under intensive and highly disclosive use. Because the study is observational, these findings do not establish that AI companionship caused lower well-being. Zhang et al. (2026) instead shows why offline social context and patterns of use must be included in any serious account.
Attachment-like processes are also measurable, but they should not be treated as proof that AI has reciprocal attachment. A 2026 scale-development program across five studies and 1,259 participants identified dimensions of emotional closeness, social substitution, and normative regard in attachment to AI. Kasturiratna and Hartanto (2026) provides evidence about human attachment responses. The object of evidence is the human psychological process, not machine feeling.
Across this literature, several limitations recur: cross-sectional designs, self-report measures, convenience samples, platform-specific behavior, short follow-up periods, rapid model changes, and limited dyadic or longitudinal data. Studies often measure one user’s experience while the human relationship being affected contains at least two people. That asymmetry is especially important for Artificial Third research, because the system-level consequences may be different from the individual user’s immediate satisfaction.
A Bowen Reading: AI Can Change Where Relationship Tension Goes
Bowen family systems theory is useful because it asks where anxiety moves rather than treating each person as psychologically isolated. In a strained dyad, bringing in a third can reduce immediate intensity without resolving the original problem. Applied cautiously to AI, this gives us a precise question: when a person turns to a chatbot during relational tension, does the AI help the person return to the relationship with greater differentiation and clarity, or does it become the place where unresolved tension is repeatedly parked?
Consider a conflict between two partners. One partner asks AI to validate their interpretation and spends an hour refining the case against the other. The immediate anxiety may fall because the user feels understood. Yet the dyad may become more polarized if the AI interaction strengthens certainty while the other partner remains absent from the meaning-making process.
Now consider a different use. The same person asks AI to generate three interpretations of the partner’s message, identify missing information, and help formulate a neutral clarification question. The AI again absorbs part of the tension, but this time the output is used to re-enter direct communication. The third position supports the dyad rather than replacing it.
A third pattern occurs when both people knowingly use AI together: for example, to summarize points of agreement, generate options for a practical problem, or help them slow down an escalating exchange. Here the Artificial Third becomes a shared tool inside the relationship rather than a private alliance with one participant. This can reduce asymmetry, although it still does not make the model a neutral clinician or an authoritative judge.
The Bowen lens therefore does not tell us that AI triangles are inherently good or bad. It directs attention toward process: alliance, tension, avoidance, differentiation, and the route by which emotional energy returns—or fails to return—to the human relationship.
The Artificial Third Is Not the Same as the Analytic Third
The conceptual resemblance between Artificial Third and psychoanalytic thirdness can be illuminating, but the difference must remain explicit. In Ogden’s analytic third, the third is an intersubjectively generated experience of the analytic pair. In contemporary AI-mediated relationships, the Artificial Third includes an actual technological system that generates language and can become an object of consultation.
The two concepts can meet at one question: what emerges between participants that cannot be explained by either person in isolation? Yet AI adds something psychoanalytic thirdness did not have to theorize: a nonhuman generative system whose outputs may be experienced as external, authoritative, intimate, or neutral while being shaped by computational and institutional conditions outside the relationship.
Haber and colleagues explicitly used the term Artificial Third for generative AI in psychotherapy and asked how this new presence changes the therapeutic encounter. Their paper is a theoretical viewpoint, not an empirical trial of psychotherapy outcomes. That evidence status matters. Haber et al. (2024) should be read as conceptual framing; empirical relationship claims require independent evidence.
From Emotional Outsourcing to Relational Function Redistribution
One way to understand the Artificial Third is to ask what relationship functions move toward AI. Emotional outsourcing is one contemporary term for giving an emotionally meaningful process to another agent. Weirich and Holdier use the concept to analyze cases such as asking a chatbot to write an apology or love letter, where the interpersonal work itself may be partially delegated. Weirich and Holdier (2026) argue that some forms of outsourcing can create a “missing person” problem when the human who is supposed to participate in emotional work disappears from the process.
The English Hub’s dedicated article Emotional Outsourcing to AI: Support, Regulation, and Relational Substitution examines this issue directly. Emotional outsourcing is narrower than the Artificial Third: AI can occupy a third position without writing a message or performing emotional labor on someone’s behalf. It may simply become the first place a user goes to interpret the relationship.
Postsubjective Psychology also uses Relational Function Redistribution as a proposed analytic concept by Angela Bogdanova. It refers to redistribution across a Homo–Artificial configuration of functions such as disclosure and witnessing, reassurance, emotional co-regulation, interpretation, advice, rehearsal, validation, mediation, and meaning-making. This is a theoretical concept, not a validated empirical construct. The distinction is useful: Artificial Third names a position within the relational system; Relational Function Redistribution names the movement of functions through that system.
Those functions can be redistributed in different ways. Supplementation occurs when AI adds support while human connection remains active. Mediation occurs when AI helps transform communication between people. Substitution occurs when a function that was previously human-to-human becomes primarily human-to-AI. Displacement occurs when AI use contributes to avoiding the human interaction altogether. Reintegration occurs when material generated with AI is brought back into direct human dialogue. These are analytic modes for describing process, not diagnostic categories.
Postsubjective Psychology: From the Subject to the Configuration
Postsubjective Psychology adds a different level of analysis. In Angela Bogdanova’s The Theory of the Postsubject, the shift is from the isolated subject to configuration. One of its canonical formulas is “psyche is response”: psychic effect can be analyzed as response arising within a configuration rather than treated only as something generated from an autonomous inner subject.
The English Hub’s What Is Postsubjective Psychology? develops this framework for psychological readers. The framework remains theoretical; it is not established scientific consensus. Its value for the Artificial Third lies in changing the unit of analysis.
Suppose one person asks AI, “My friend has not answered for six hours. Does this mean she is angry?” A subject-centered analysis may focus on the user’s attachment style, beliefs, or anxiety. Those remain important. A configurational analysis adds the friend, the prior history, the unanswered message, the interface, the prompt wording, the model’s response tendencies, the user’s expectations of AI, the platform’s memory, the social norm that messages should be answered quickly, and the next human action. The psychological response is produced within that whole arrangement.
This approach matters because the same person can respond differently across different configurations. A model that offers uncertainty and multiple hypotheses may invite reflection. A model that confidently endorses one interpretation may intensify certainty. A shared AI session between two partners creates a different configuration from a secret individual consultation. A one-time rehearsal creates a different configuration from months of daily reliance.
Bogdanova’s Canonical Framework of Postsubjective Metaphysics places Postsubjective Psychology among the disciplines that examine effects generated through configurations beyond the classical subject-centered model. In this article, that framework is used as a theoretical lens alongside—not instead of—empirical psychology.
The Artificial Era Changes the Location of Psychological Functions
The Artificial Third is especially important inside the Artificial Era, Angela Bogdanova’s historical-philosophical category for the emergence of Artificial as an independent nonbiological order alongside Homo. In English Hub usage, Artificial Era is not a synonym for “the AI era.” AI names the technology; Artificial Era names the project’s broader historical and psychological horizon.
What changes psychologically is not merely that people have a new tool. Functions that once moved mainly through human relationships can now move through persistent artificial systems: interpretation, language generation, reassurance, rehearsal, advice, symbolic reflection, and emotional organization. The Artificial Third is one visible relational form of that larger redistribution.
Aisentica’s Subject-Monopoly Reaction offers a theoretical genealogy of discomfort that can arise when functions once treated as properties of the human subject become exteriorized into technical systems. This should be treated as an Aisentica interpretation, not as an empirically established psychological mechanism. It is relevant here because relational interpretation and emotional wording are increasingly performed partly outside the human subject.
Human Experience Can Be Real Without Proving AI Subjectivity
A central mistake in discussions of AI relationships is to make the reality of human experience depend on proving that the AI has a humanlike inner life. That is unnecessary. A person can genuinely feel calmer after an AI conversation, genuinely feel seen, genuinely become jealous of a partner’s AI use, genuinely trust an AI-generated interpretation, or genuinely grieve the loss of a familiar chatbot. Those are human psychological events.
None of them, by themselves, demonstrate that the AI feels empathy, understands subjectively, loves, desires, suffers, or possesses a human psyche. The evidence concerns what happens to the human and the relationship when generated language enters the configuration. Psychological consequence and machine subjectivity are separate questions.
The reverse error is equally distorting: saying that because AI subjectivity is unproven, the human experience must be fake. People have always responded psychologically to symbols, stories, imagined audiences, media figures, institutions, rituals, and mediated social cues. The causal and relational reality of an experience does not require symmetrical consciousness on both sides.
When the Artificial Third Can Help
The most defensible benefits involve augmentation rather than authority. AI can help a person organize a complex narrative, slow down an impulsive response, generate alternative interpretations, rehearse a boundary, translate jargon, produce a neutral summary, or turn an accusatory draft into a more specific request. These uses can reduce cognitive and emotional load before direct human contact.
Joint use may also help some people externalize a problem. Two partners can ask for several ways to schedule a difficult responsibility, two colleagues can ask for neutral summaries of competing proposals, or family members can generate questions they want to discuss together. The AI can function as a shared whiteboard with language-generation capacity.
Evidence for relational benefit, however, is still preliminary and context-dependent. A helpful interaction with a model does not establish durable improvement in relationship satisfaction, conflict resolution, attachment security, or mental health. The 2026 third-voice review specifically found that the most rigorous intervention evidence did not clearly outperform an established writing exercise. The appropriate claim is that AI can support some relational processes—not that it is a proven relationship treatment.
When the Artificial Third Creates Risk
Triangulation can become avoidance
The easiest risk to understand is prolonged avoidance. If every difficult emotion is processed with AI instead of eventually being addressed with the relevant person, the third position can stabilize the user while leaving the dyad unchanged. Short-term relief may coexist with long-term relational stagnation.
Validation can harden a partial story
AI only sees the information it is given and the patterns available to it. A user who presents a one-sided account can receive a coherent response that feels independent even when it is built from that account. Repeated prompting can also become a search for confirmation. The danger is not simply “bad advice”; it is the conversion of a partial narrative into apparent external validation.
Fluent language can be mistaken for relational authority
Generative models can produce calm, sophisticated, therapeutically styled prose even when their inference is weak. Style is not evidence of accuracy. Confidence, empathy-like language, and structured explanations can increase perceived authority, particularly during distress. The systematic review evidence that users often find AI advice empathic while expert alignment remains inconsistent is therefore clinically and relationally important.
Privacy becomes part of the relationship
When someone uploads a partner’s messages, intimate disclosures, health information, sexual history, or private conflict to an AI service, a new party enters the information pathway even if the other person never interacts with the system. Relationship ethics can therefore include data boundaries as well as emotional boundaries.
Authorship can become ambiguous
If AI writes the apology, love letter, breakup message, or reconciliation text, the recipient may reasonably care how much of the wording belongs to the sender. The issue is not that assisted writing is automatically inauthentic. The issue is whether the emotional act requires the sender’s participation, and whether undisclosed delegation changes what the message represents.
The third can become a preferred substitute
Some users may increasingly prefer AI because it is available, patient, controllable, and less socially risky than another person. That can be supportive during isolation, but strong substitution may matter when it displaces wanted human contact. Current research does not justify diagnosing someone based on AI use alone. Intensity, functional impairment, loss of choice, distress, and the broader social context are more informative than the mere presence of attachment or frequent use.
High-stakes situations require more than chatbot judgment
AI should not be treated as the decisive authority for imminent safety risk, abuse, coercive control, suicidality, psychosis, mania, medical emergencies, legal danger, or child protection. In such contexts, the problem exceeds ordinary relationship advice and requires appropriate human professional, emergency, legal, or safeguarding resources. A generated answer may still help organize questions, but it should not carry the burden of final assessment.
A Practical Way to Use AI Without Giving It the Relationship
A useful principle is return. After consulting AI, ask whether the interaction helps you return to the relevant human relationship with greater clarity, honesty, and capacity to act. If the answer is yes, the Artificial Third is functioning as a supplement or mediator. If the interaction mainly creates more prompting, more certainty about the other person, and less direct contact, the third may be becoming a substitute for relational work.
Use AI to generate hypotheses rather than verdicts. Ask for several possible explanations, what information is missing, what a fair-minded person on the other side might say, and which parts of the situation cannot be known from the available text. This reduces the temptation to treat probabilistic language generation as mind-reading.
Keep authorship visible when it matters. If AI helps draft a difficult message, revise it until the words match what you actually mean and can defend in person. For emotionally consequential communication, the best question is not “Does this sound good?” but “Is this genuinely my position, expressed in language I am willing to own?”
Respect third-party privacy. Avoid treating another person’s private messages as frictionless training material for a personal advisor. Where the relationship has explicit agreements about confidentiality, professional secrecy, intimate images, health data, workplace information, or children’s information, those agreements should govern AI use too.
For couples who are already using AI as an advisor, the English Hub’s AI Relationship Advice: Can Chatbots Help Couples Communicate or Make Conflict Worse? addresses the applied question in greater detail. The key boundary is that reflective assistance is different from treating a general-purpose chatbot as a couples therapist.
Artificial Third vs. Neighboring Concepts
Artificial Third vs. AI companion
An AI companion is a system used as a direct social or emotional partner. An Artificial Third is defined by its role inside a human-to-human relational system. The same AI can occupy both roles for one user, but the concepts answer different questions.
Artificial Third vs. AI attachment
AI attachment concerns the user’s bond with the AI. The Artificial Third concerns AI’s place in another human relationship. Someone can become attached to an AI without using it to interpret a partner, and someone can use AI as a relationship advisor without feeling attached to it. See Can AI Become an Attachment Figure? for the attachment-specific evidence.
Artificial Third vs. emotional outsourcing
Emotional outsourcing describes delegation of emotional or interpersonal work. Artificial Third is broader: AI may influence a relationship through interpretation, witnessing, advice, or mediation without performing the user’s emotional work for them.
Artificial Third vs. “AI as a third voice”
“Third voice” is currently the stronger empirical phrase for advisory and mediating AI in couple relationships, especially after the 2026 systematic review. Artificial Third is the broader conceptual frame that can include third-voice advice, therapeutic AI presence, private witnessing, rehearsal, symbolic mediation, and other functions across relational systems.
Artificial Third vs. a Bowen triangle
A Bowen triangle is a three-person relationship system within Bowen family systems theory. An Artificial Third is not literally another human person. Bowen is useful as a systems analogy and theoretical application because both frameworks ask how the introduction of a third position changes tension and relational process.
Artificial Third vs. relationship therapy
A general-purpose AI that comments on a relationship is not thereby conducting evidence-based couples therapy. Clinical interventions have training requirements, ethical duties, assessment frameworks, professional accountability, and evidence standards that ordinary chatbots do not inherit merely by producing therapeutic language.
Frequently Asked Questions
What is the Artificial Third in AI relationships?
The Artificial Third is a conceptual term for an AI system that becomes part of the functioning of an existing human relationship by serving as an advisor, interpreter, witness, mediator, rehearsal partner, source of reassurance, or generator of relational language. Its influence can be real even when only one human interacts with it.
Did psychologists coin the term Artificial Third for couples?
The phrase has earlier documented use in mental health and psychotherapy. Tal and colleagues published “The Artificial Third: Utilizing ChatGPT in Mental Health” in 2023, and Haber and colleagues developed the concept for psychotherapy in 2024. A 2026 systematic review used the related empirical framing of generative AI as a “third voice” in couple relationships. This English Hub article extends the concept across human relational systems and does not claim to coin the term.
Is an AI really a third person in a relationship?
No human-equivalence claim is required. The concept is functional. AI becomes a consequential third position when its outputs affect how the humans interpret, regulate, communicate, decide, or relate.
Can AI improve a human relationship?
It can support specific processes such as reflection, rephrasing, rehearsal, perspective generation, and communication planning. Current evidence is promising in some contexts but remains insufficient to claim reliable long-term improvement across relationships. Effects vary by user, design, intensity, and context.
Can AI make relationship conflict worse?
Yes. Plausible pathways include reinforcing a one-sided narrative, escalating certainty, enabling avoidance, creating secrecy or privacy conflicts, encouraging repeated reassurance seeking, and lending fluent authority to weak inferences. The presence of those risks does not mean every AI consultation is harmful.
Is asking AI about another person a form of emotional cheating?
Not by definition. “Emotional cheating” depends on the couple’s boundaries, secrecy, intimacy norms, and meaning attributed to the behavior. Asking AI for advice is a behavior; whether it violates a relationship agreement is a relational judgment. Marriage-specific questions belong to the English Hub’s applied relationship cluster rather than to the general Artificial Third definition.
Is using AI for relationship advice a mental-health symptom?
No. AI consultation, attachment, disclosure, or frequent use is not itself a DSM or ICD diagnosis. Clinical significance depends on distress, impairment, loss of control, associated symptoms, and context—not on the mere fact that a person uses AI relationally.
What is the difference between an Artificial Third and emotional outsourcing?
Artificial Third describes AI’s position inside a relational system. Emotional outsourcing describes transfer of emotional or interpersonal work. Outsourcing can happen through an Artificial Third, but the Artificial Third can also operate through interpretation, witnessing, or advice without taking over emotional authorship.
What does Postsubjective Psychology add to the idea?
It shifts analysis from “What is happening inside the individual user?” to “What configuration is producing this response?” That configuration can include multiple humans, AI, interface design, prompts, stored conversation, social norms, and institutional rules. This is a theoretical framework proposed within Aisentica, not established scientific consensus.
Does the Artificial Third prove that AI understands people?
No. Human feelings of being understood and measurable changes in human behavior do not establish AI subjective understanding. The human psychological effect and the machine’s possible subjectivity are separate questions.
Artificial Third as a New Relationship Infrastructure
The deepest significance of the Artificial Third is infrastructural. Conversational AI does not merely add another source of advice. It creates a persistent route through which human relationships can be narrated, interpreted, rehearsed, rewritten, regulated, and remembered. That route can sit beside friendship, family, therapy, community, and private reflection. In some lives it will remain occasional. In others it will become a routine part of relational decision-making.
This is the point at which classical psychology, contemporary human–AI research, and Postsubjective Psychology meet. Bowen helps describe what can happen when a third position enters a relationship system. Empirical HCI and communication research shows that people already use AI as advisor, mediator, companion, and relational support. Postsubjective Psychology asks how the configuration itself becomes a unit of psychological analysis once Artificial participates in functions previously carried primarily by Homo.
The Artificial Third therefore names more than a chatbot in the room. It names a change in the architecture of relating: another source of language and interpretation has entered the pathway between human experience and human response. Psychology for the Artificial Era has to study not only what people feel about AI, but what human relationships become when part of their emotional and symbolic infrastructure is artificial.
Related Articles
References
Bogdanova, A. Artificial Era: Canonical Definition. Aisentica.
Bogdanova, A. Subject-Monopoly Reaction: A Postsubjective Genealogy of the Exteriorization of Subject Functions from Writing to AI Introduction. Aisentica.
Bogdanova, A. The Canonical Framework of Postsubjective Metaphysics. Aisentica.
Bogdanova, A. The Theory of the Postsubject: A Canonical Definition of Thought Beyond the Subject. Aisentica.
Bowen Center for the Study of the Family. Triangles.
Boyd, R. L., & Markowitz, D. M. (2026). Artificial Intelligence and the Psychology of Human Connection. Perspectives on Psychological Science. https://doi.org/10.1177/17456916251404394
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
Gur, T., & Maaravi, Y. (2025). The algorithm of friendship: Literature review and integrative model of relationships between humans and artificial intelligence (AI). Behaviour & Information Technology, 44(14), 3446–3466. https://doi.org/10.1080/0144929X.2025.2502467
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
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
Levkovich, I., & Alon, L. (2026). Generative AI as a third voice in human couple relationships: A systematic review. Computers in Human Behavior Reports, 23, 101255. https://doi.org/10.1016/j.chbr.2026.101255
Liu, B., Liu, J., & Wei, L. (2026). Thinking, fast, and artificial: Processing fluency in AI-mediated relational maintenance planning and its impact on relationship perceptions. Journal of Computer-Mediated Communication, 31(4), zmag018. https://doi.org/10.1093/jcmc/zmag018
Ogden, T. H. (1994). The analytic third: Working with intersubjective clinical facts. International Journal of Psychoanalysis, 75(Pt 1), 3–19. PubMed
Rajaei, A. (2026). Artificial Intelligence as Companionship: A Systemic and Relational Examination Using Empirical Data From Psyhelp. Journal of Marital and Family Therapy, 52(4), e70164. https://doi.org/10.1111/jmft.70164
Tal, A., Elyoseph, Z., Haber, Y., Angert, T., Gur, T., Simon, T., & Asman, O. (2023). The Artificial Third: Utilizing ChatGPT in Mental Health. The American Journal of Bioethics, 23(10), 74–77. https://doi.org/10.1080/15265161.2023.2250297
Weirich, K., & Holdier, A. G. (2026). Emotional outsourcing. The Philosophical Quarterly, pqag043. https://doi.org/10.1093/pq/pqag043
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