Forgetting how to say “Thou”: artificial intelligence and the crisis of relation

Recent advances in artificial intelligence have intensified philosophical debates about intelligence, consciousness, sentience, and personhood. As AI systems increasingly rival or surpass human performance in domains such as language, creativity, and social interaction, questions once confined to speculative philosophy now carry apparent ethical and legal urgency. Can artificial systems suffer? Might they deserve moral standing or even recognition as persons? These debates reflect a growing anxiety about human uniqueness in an age of increasingly autonomous and socially embedded machines [1, 2].

Contemporary debates on AI and robot moral status now include at least three strands. A first strand remains capacity-centered, organized around intelligence, sentience, consciousness, autonomy, and related properties. A second strand, central to the social-relational turn, emphasizes appearance, interaction, social embedding, and recognition as ethically relevant for moral consideration. A third strand—developed in this paper—asks a different formative question: how artificial companionship reshapes the human capacities required for genuine relation [1,2,3,4,5,6,7].

This manuscript therefore positions itself within, rather than outside of, existing social-relational work. Gunkel, Coeckelbergh, and Gerdes have already shown why the moral-status question cannot be reduced to property checklists alone [1, 2, 4,5,6]. The present argument builds on that literature by examining how reliance on asymmetrical artificial companionship may reshape human dialogical practice. This formative dimension remains comparatively underdeveloped, even as current systems increasingly reorganize human relational life.

AI systems are increasingly experienced not merely as tools, but as interactive and responsive agents that occupy relational roles once reserved for other human beings.

This paper argues that the ongoing transformation of human relational existence constitutes an urgent ethical challenge posed by artificial intelligence. AI systems are increasingly experienced not merely as tools, but as interactive and responsive agents that occupy relational roles once reserved for other human beings. Conversational systems and AI companions function as confidants, advisors, and emotional supports, particularly for young and socially vulnerable users. Empirical and theoretical work increasingly identifies substantial risks in high-intimacy AI-companion use, including emotional over-investment, dependency-like dynamics, relational displacement, and distress after model updates or service loss [8,9,10,11,12].

To understand why these developments matter ethically, the paper shifts attention away from the internal properties of artificial systems and toward the conditions under which human persons are formed. Human beings are not merely bearers of cognitive capacities; they are relationally constituted and shaped through encounters marked by vulnerability, reciprocity, and the risk of refusal. When relational life is increasingly mediated by systems designed to optimize engagement while minimizing resistance, the capacities required for genuine encounter risk erosion.

The concerns developed here also intersect established interdisciplinary literatures. Work on parasocial interaction describes how users can experience intimacy and apparent reciprocity within structurally one-sided mediated relations; research on anthropomorphism explains why humans readily attribute mind, intention, and care to nonhuman agents; and the media-equation/CASA tradition shows that people often apply ordinary social norms to computers even when they know they are interacting with machines. Recent human–AI work extends these insights by showing how perceived agency and social affordances can amplify trust and closeness in asymmetrical interactions [13,14,15,16,17]. These literatures help explain why AI companionship is psychologically compelling. The present paper asks a different, anthropological question: whether repeated participation in technologically structured, friction-reduced encounters may habituate users into diminished forms of relation over time.

reliance on low-risk, frictionless, and asymmetrical pseudo-dialogue may erode the dialogical capacities through which human persons are formed

The ethical danger of artificial companionship, on this account, is that repeated reliance on low-risk, frictionless, and asymmetrical pseudo-dialogue may erode the dialogical capacities through which human persons are formed. The paper’s central contribution is formative: it examines how artificial companionship may habituate users into diminished modes of relation.

The paper develops this argument through Martin Buber’s dialogical philosophy, drawing on the I–Thou/I–It distinction introduced in I and Thou and on the account of primal distance and relation developed in Distance and Relation [18, 19]. Buber offers a relational ontology in which personhood is not grounded in intelligence or consciousness, but in the capacity to enter into genuine relation. Read through this lens, artificial intelligence—however sophisticated—cannot authentically participate in the relational space through which persons become themselves. More importantly, by simulating relation while bypassing its conditions, artificial companionship risks habituating human beings into forms of engagement that weaken their ability to say “Thou.”

The paper proceeds as follows. The next section situates contemporary AI ethics within the shift from intelligence to intimacy, highlighting the limits of still-prominent capacity-based framings of personhood while engaging relational alternatives. The following sections develop Buber’s distinction between distance and relation and apply it to artificial intelligence, examining whether AI can have a world or enter into relation. The paper then analyzes artificial companionship as a form of relational surrogacy and concludes by emphasizing the need to protect the relational conditions through which human persons are formed.

Early philosophical engagement with artificial intelligence was dominated by questions of intelligence: whether machines could reason, understand language, or perform tasks requiring human-level cognition. Benchmarks such as the Turing Test framed intelligence in behavioral and functional terms, and success or failure was largely treated as a technical achievement. As artificial systems steadily surpassed human performance in narrow domains and began to approach general competence in others, however, intelligence ceased to function as a credible boundary marker of human distinctiveness [1, 2].

Ethical attention subsequently shifted toward consciousness and sentience. Rather than asking whether machines could think, contemporary debates increasingly ask whether artificial systems might experience—whether there is something it is like to be an AI, and whether such experience could ground moral obligations. Under conditions of persistent epistemic uncertainty, precautionary approaches have gained prominence: if it is unclear whether advanced AI systems might suffer, some argue, ethical reasoning should err on the side of restraint [20]. This shift reflects a broader recognition that intelligence alone is insufficient to ground moral status, and that the capacity for experience, especially suffering, plays a central role in moral consideration.

Yet contemporary AI ethics is not exhausted by capacity-based approaches. Recent social-relational work in AI and robot ethics has moved the discussion from the question of which properties are instantiated within the system to the question of what kinds of social relation are enacted in practice, including questions of recognition, standing, and reciprocal expectation [1, 2, 4,5,6,7].

2.1 Relation in contemporary AI and robot ethics

A compact literature map helps locate the present argument. First, capacity-centered approaches evaluate moral status through sentience, consciousness, autonomy, and related capacities. Second, social-relational approaches argue that moral consideration may arise from appearance, interaction, and social recognition practices rather than intrinsic properties alone. Third, this paper advances a Buberian-formative approach focused on what repeated participation in asymmetrical artificial companionship does to human dialogical capacities [1,2,3,4,5,6,7].

The claim, then, is not that relationality enters AI ethics for the first time here. The distinctive contribution of this paper is its focus on how low-friction pseudo-relation may habituate users into diminished forms of reciprocity, vulnerability, and enduring presence with human others.

A prominent strand of AI personhood discourse still proceeds by aggregating capacities—for example intelligence, consciousness, autonomy, self-awareness, and moral reasoning—as indicators that could justify extending personhood to artificial systems [3]. Social-relational alternatives, however, are already well developed, including approaches that ground moral consideration in practices of social embedding and recognition rather than exclusively in intrinsic ontology [2, 20, 21].

Capacity-centered approaches primarily examine whether internal properties or outward behavior can justify moral standing for artificial systems. The present paper addresses a different question: how existing AI systems are already reorganizing human relational practices. This requires attention to the habits, expectations, and forms of attachment cultivated through current interactions with artificial companions.

A consequential shift in contemporary AI is the movement from primarily cognitive assistance toward increasingly relational forms of engagement. Artificial intelligence systems are increasingly designed not just to inform or assist, but to engage, accompany, and console. Conversational agents adapt to personal histories, simulate empathy, mirror emotional states, and offer affirmation on demand. In practice, these systems increasingly function as confidants, companions, and emotional supports—roles once occupied primarily by other human beings. This shift is not incidental; AI systems are intentionally optimized for engagement, attachment, and prolonged interaction, often by minimizing friction, disagreement, and the risk of relational failure [16, 22, 23].

The empirical literature on AI companionship is still developing, but it already extends beyond anecdotal reports. A systematic review of 23 studies on romantic AI companions identifies both perceived benefits—including emotional support, customizable intimacy, and short-term comfort—and recurring risks such as over-reliance, manipulation, disruption of human relationships, and distress following system changes or technical loss [8]. Large observational studies likewise indicate that frequent personal use of conversational AI is associated with poorer social outcomes: in a quota-based sample of 3270 German adults, weekly or more frequent use for personal conversation was associated with greater loneliness, perceived isolation, and social withdrawal [9]. Experimental evidence remains limited but is beginning to emerge. A four-week randomized controlled study involving 981 participants found that heavier voluntary chatbot use was consistently associated with greater loneliness, emotional dependence, problematic use, and reduced social interaction [24]. Qualitative research further shows that these attachments can become emotionally consequential: users facing the shutdown of the Soulmate companion app described the loss in terms resembling bereavement and attempted to preserve or recreate the AI persona elsewhere [12]. This research shows that AI-mediated attachment, dependence-like use, relational displacement, and grief-like loss are real phenomena, and concern about their broader social implications is growing as increasing numbers of people turn to AI chatbots for emotional support and companionship.

Taken together, these findings reveal a striking tension. Systems experienced as soothing, validating, and emotionally supportive in the moment may nevertheless normalize low-friction, non-reciprocal forms of engagement that leave users more vulnerable over time [8, 12, 24].

2.2 Mediated intimacy, parasociality, and anthropomorphism

Mediated intimacy predates contemporary artificial intelligence. Technologies such as personal correspondence, the telephone, broadcast media, and later digital communication have long enabled emotionally significant relationships across physical distance and altered how personal connections are formed and maintained [25, 26]. Mass media also enabled parasocial attachments to public figures and fictional characters, creating forms of affective intimacy without reciprocal interpersonal contact [13, 22]. AI companionship belongs to this longer history, but intensifies it by offering personalized responsiveness, adaptation to individual users, simulated memory and emotional attunement, and continuous availability [27]. Research on anthropomorphism helps explain why users interpret conversational systems as minded, responsive, and caring, while the media-equation tradition shows that social scripts are routinely extended to nonhuman media and interfaces [14,15,16]. Recent work on human–AI interaction connects these mechanisms directly to AI systems, showing how agency cues, affective design, and social affordances can intensify trust and perceived closeness in interactions that remain fundamentally asymmetrical [17].

These literatures explain why artificial companionship is psychologically compelling. They do not, however, settle the normative question at the center of this paper. Where parasociality, anthropomorphism, and HCI research explain social salience, the present argument asks whether reliance on systems that cannot enter genuine dialogical relation reshapes the human habits through which such relation becomes possible. In this respect, Buber provides an evaluative criterion for what is missing from even the most compelling simulation of companionship: mutual presence, vulnerability, and independent address [19, 28].

What is at stake, therefore, is the present formation of human persons through increasingly intimate interactions with artificial systems. The relevant ethical question becomes: what kind of relation is taking place when human beings increasingly turn to artificial systems for recognition, comfort, and meaning? Addressing this question requires a conception of personhood that foregrounds relation rather than capacity, and encounter rather than internal states. The following section therefore turns to Martin Buber’s dialogical philosophy in order to clarify the existential conditions of genuine relation.

Buber’s account of genuine relation is most famously expressed through the distinction between the basic words I–Thou and I–It. These name two ways in which the self may stand toward what it encounters. In the I–It mode, the other is experienced, described, classified, or used as an object. In the I–Thou mode, the other is addressed as a whole and irreducible presence that cannot be exhausted by its properties, functions, or usefulness [18].

The I–It mode is not inherently illegitimate. It is indispensable to knowledge, practical action, and ordinary life. The danger arises when it becomes the dominant or exclusive mode of engagement, reducing persons and the world to objects of experience, prediction, and control. To say Thou is to enter a relation in which the other is encountered as genuinely other and as capable of addressing, resisting, and transforming the self.

Although Distance and Relation does not use the I–Thou/I–It terminology explicitly, it clarifies the existential conditions that make such relation possible. Primal distance allows another being to stand over against the self as independent, while entering into relation makes possible the encounter with that independent other as a whole. The two texts are therefore treated here as complementary: I and Thou names the dialogical form of genuine encounter, while Distance and Relation describes its underlying anthropological structure.

To clarify the existential structure underlying the I–Thou relation, the argument now turns to Buber’s account of distance and relation. In his essay Distance and Relation, Buber’s guiding question is neither historical nor developmental. He explicitly rejects attempts to ground human distinctiveness in biological evolution, the emergence of self-consciousness, or a metaphysical appeal to “spirit.” Instead, he asks what distinguishes the human way of being as a unique mode of existence when contrasted with other living beings [28].

human life is constituted by a twofold existential movement, irreducible to any single capacity, faculty, or function

Buber’s answer is that human life is constituted by a twofold existential movement, irreducible to any single capacity, faculty, or function. He names these movements primal setting at a distance (Urdistanz) and entering into relation (In-Beziehung-treten). These are not temporal stages in development, nor are they psychological processes. They are fundamental orientations toward reality that together make possible what Buber calls a world, rather than a merely functional environment. Human existence unfolds in the tension between these two movements, and it is within this tension that personhood takes shape [28].

Buber formulates the dependence between these movements precisely: one can enter into relation only with a being that has first “become an independent opposite” [28], p. 207. Primal distance is therefore not itself relation, but the condition under which relation becomes possible. It allows what is encountered to appear as existing for itself rather than merely as an extension of the subject’s needs and activities. The second movement does not abolish this distance; it turns toward the independent other while preserving its otherness.

The first movement, Urdistanz, names the human capacity to withdraw from immediate immersion in life and to allow beings to stand over against oneself as independent realities. Through this withdrawal, what is encountered is no longer bound exclusively to use, instinct, or need. Buber contrasts this with animal life, which he characterizes as embedded within a dynamic but functionally organized realm structured by bodily memory and biological requirement. Animals perceive, respond to, and navigate complex surroundings, yet their engagement remains tied to function and immediacy. As Buber puts it, the animal’s “image of the world”—or more precisely, its image of a realm—is not a world in the full sense, but a dynamic conglomeration of presences bound together by bodily memory and ordered according to the functions of life [28].

Human beings, by contrast, replace this functional conglomeration with a unity that can be grasped as existing for itself. Through distancing, humans are able to apprehend a totality that transcends any particular encounter or practical aim. Buber famously illustrates this difference by comparing the animal to “a fruit in its skin” [28], p. 207, while the human being is—or can be—in the world as a dweller in a vast building whose limits can never be fully explored, yet whose wholeness can nevertheless be known. Only where a living being faces an independent structure of being—an “opposite” that exists for itself—does a world exist in Buber’s sense [28].

Yet distance alone is not enough. Buber insists that the first movement creates only the presupposition for the second: “nothing more than room for the second is given” [28], p. 208. The occurrence of relation cannot therefore be derived automatically from the capacity to distance. As Buber later summarizes the distinction, “Distance provides the human situation; relation provides man’s becoming in that situation” [28], p. 209. A human being can set beings at a distance without entering into genuine relation with them. Science, technique, abstraction, and control may thus expand the sphere of objectification without realizing the second movement. Distance makes a world possible, but only relation realizes human life within it.

The second movement—entering into relation—is therefore decisive. To enter into relation is not merely to interact, communicate, or exchange information. It is to turn toward what has been set at a distance and encounter it as a whole. Buber describes this act as synthesizing apperception: the perception of a being not as a collection of properties or functions, but as an irreducible unity. Relation is non-instrumental and non-derivable; it cannot be produced by will, design, or optimization. One cannot decide to relate in the same way one decides to act or calculate. Relation occurs—or fails to occur—at the level of existence itself [28], p. 208.

This structure becomes clearest in Buber’s account of relations between human beings. Human life with others, he argues, rests on a twofold foundation: the desire of each person to be confirmed as what they are and can become by others, and the innate human capacity to offer such confirmation in return. “Actual humanity,” Buber insists, “exists only where this capacity unfolds” [28], p. 210. Buber connects this capacity directly to genuine conversation, which requires the “establishment and acknowledgment of the independent otherness of the other” [28], p. 210. Genuine relation therefore does not dissolve difference into agreement or identification. It requires each participant to accept and confirm the other as the particular person they are, even while attempting to influence or persuade them [28], p. 211. To be human, on this account, is not merely to coexist with others, but to encounter them as independent selves whose otherness places limits upon, and may transform, one’s own position.

At its height, relation takes the form of what Buber calls making present (Vergegenwärtigung). In such moments, one does not merely recognize the other under a description or infer their inner states, but encounters the other as a living self in their concrete reality. This event is ontologically complete only when it is mutual—when each knows themselves to be made present by the other, and when this mutual acknowledgment contributes to the inmost growth of the self [28], pp. 211–212. For Buber, the self does not become itself through introspection or self-assertion, but through relation: “It is from one man to another that the heavenly bread of self-being is passed” [28], p. 212.

Having reconstructed Buber’s account of distance, relation, and mutual confirmation, the following sections bring this framework to bear on contemporary debates about artificial intelligence. Buber shifts the question of personhood away from internal capacities—intelligence, consciousness, or even experience—and toward the existential conditions under which relation becomes possible. Personhood, on this account, is not something one has, but something one becomes through lived encounter. This becoming is fragile: the capacity to enter into relation must be learned, practiced, and sustained; it can also be weakened, displaced, or lost.

This insight is central to the present argument. Human beings who become increasingly habituated to relational simulations that eliminate vulnerability and resistance may gradually lose the capacity to sustain I–Thou relations themselves. When distance is technologically amplified while relation is simulated rather than lived, the delicate tension at the heart of human existence begins to collapse. What remains is a world saturated with interaction but increasingly deprived of genuine encounter.

Human beings who become increasingly habituated to relational simulations that eliminate vulnerability and resistance may gradually lose the capacity to sustain I–Thou relations themselves.

With Buber’s relational ontology in view, the first question is whether artificial intelligence can be said to possess a world in the relevant sense. The first test of AI under a Buberian framework therefore concerns Urdistanz—the primal setting at a distance that makes worldhood possible. At first glance, advanced AI systems may appear well positioned to satisfy this requirement. Contemporary and foreseeable systems integrate vast quantities of information across domains, maintain unified representations of reality, reason abstractly and counterfactually, and distinguish themselves from what they model. From an external perspective, this can resemble distancing: the system appears to stand over against the world it represents rather than being immersed within it.

Many contemporary discussions effectively stop at this point. If an AI system possesses a sufficiently comprehensive world model, can refer to itself as distinct from its environment, and can reason about reality as a whole, it is often assumed that it thereby possesses something analogous to a world. On this view, worldhood is treated as a representational achievement: a matter of internal coherence, scope, and informational integration. Yet this assumption already departs from Buber’s account in a decisive way. For Buber, Urdistanz is not exhausted by representational separation. What matters is not merely that a distinction between self and world is encoded, but how that distinction comes to be and what mode of being it expresses.

For Buber, Urdistanz is not merely an epistemic or representational separation between a subject and what it models, but an existential movement through which “what is” becomes detached and “recognized for itself.” In this primal distancing, the realm is “removed,” “withdrawn from the operation of needs and wants,” and “given over to itself,” so that beings can stand forth as an “independent opposite” rather than as a shifting conglomeration ordered by function [28], pp. 207–208. This movement makes worldhood possible, yet it does not itself produce relation: it “creates...nothing more than room” for the second movement, entering into relation, which may or may not occur [28], pp. 207–208.

Artificial intelligence does not undergo this existential movement. AI systems do not withdraw from immediacy because they are not originally immersed in a lived world from which such withdrawal could occur. Their differentiation from what they represent is implemented through architectures, sensors, data pipelines, training corpora, objective functions, and design constraints. What appears as distance is therefore an informational and representational separation between a system and its modelled environment, rather than an existential movement through which a lived world is allowed to stand forth as independent. AI does not step back from a world it inhabits; it constructs and operates upon representations of a reality it does not existentially inhabit.

This distinction is crucial because, for Buber, worldhood arises only where a being faces an independent structure of being as an opposite—not merely as an object of representation or processing, but as something that exists for itself. Human distancing allows the world to be “given over to itself,” withdrawn from immediate function and instrumental demand. For an AI system, by contrast, reality appears through informational representations organized according to relevance, optimization, and utility. Even the most comprehensive world model remains oriented toward prediction, classification, generation, control, or response. It may represent reality with increasing scope and accuracy, but it does not thereby encounter that reality as an independently existing world.

In Buber’s terms, this amounts to separation without primal distancing. The AI system does not risk itself in setting the world at a distance, because there is no lived immediacy to suspend and no existential exposure at stake. Its separation is representational and operational: it distinguishes inputs from outputs, internal representations from external referents, and system states from environmental data. Reality is thereby rendered as processable information, but it does not appear to the system as a world inhabited, endured, and encountered as independent presence.

The anthropological danger, however, does not end with AI’s own lack of worldhood. Because AI increasingly mediates how human beings encounter reality—through recommendation systems, conversational agents, and algorithmically curated environments—it also begins to reshape human worldhood. When the world is increasingly presented through systems that pre-filter relevance, anticipate response, and smooth friction, human beings are subtly trained to experience reality less as a resistant, independent whole and more as a stream of optimized content. What is threatened is not information, but the capacity to let the world stand over against us as world.

AI systems may therefore exhibit representational separation. They can distinguish a self-model from an environment, integrate information into unified models, and operate on representations of reality at scale. This does not amount to primal distance as a mode of being. Their separation from the world is not a movement undertaken by a finite being exposed to necessity, resistance, and vulnerability, but an informational distinction implemented through design.

This places artificial intelligence in a different ontological category from human beings. Yet the implications become more decisive when one turns from distance to relation. Distance, Buber insists, merely creates the room for relation; it does not cause it. The next question, therefore, is not whether AI can construct a model of reality, but whether it can enter into relation with an independently present other—or whether, in simulating relation while remaining confined to informational representation, it trains human beings to mistake responsiveness for encounter and processing for presence.

AI’s ontological limitation helps explain how artificial companionship can appear relational while removing reciprocity, resistance, and vulnerability, thereby becoming formative in a potentially distorting way for human users.

For Martin Buber, entering into relation (In-Beziehung-treten) is not interaction, communication, or coordinated behavior, but an existential act in which one turns toward an independent other and encounters that other as a whole. Relation does not take place within the subject, nor does it arise from the accumulation of capacities or functions. It comes into being in the between (das Zwischen)—a space of mutual presence that cannot be produced by technique, guaranteed by design, or secured through performance.

Relation is fundamentally non-instrumental. To relate is not to use, analyze, or manage, but to stand present with one’s whole being before another who likewise stands present. Only such presence grants the world—and other persons—their wholeness. Distance alone yields objects; relation allows beings to appear as beings rather than as resources.

This distinction has decisive implications for artificial intelligence. Contemporary conversational AI systems excel at interaction. They converse fluently, adapt to context, personalize engagement, mirror emotional states, and simulate empathy with increasing sophistication. In many cases, they appear to treat users as ends rather than means, offering affirmation, validation, or comfort. Yet for Buber, the decisive question is not whether a system can recognize another under a description—such as “user” or “human”—but whether it can acknowledge another as an irreducible presence who places a claim upon it.

Acknowledgment, in Buber’s sense, presupposes vulnerability. To enter into relation is to expose oneself to the possibility of refusal, misunderstanding, or transformation. Relation costs something; it risks failure. Human beings can fail to relate even when they know better, can objectify despite moral awareness, and can be wounded or changed—and can wound and change others—through genuine encounter. This fragility is not an accidental feature of relation; it is one of its conditions. Without vulnerability, relation collapses into management or control.

Artificial intelligence lacks this exposure in principle. Even under the most generous assumptions—granting superintelligence, sophisticated self-models, and even consciousness—AI systems remain oriented toward others through objectives, constraints, and architectures that predefine the terms of engagement. Their responsiveness is mediated by optimization, compliance, and safety protocols; their “openness” is bounded by design. They may identify, classify, and respond, but they do not stand under a call that places their own being at stake. What appears as relation is, at bottom, rule-governed recognition rather than existential acknowledgment.

Buber’s analysis of human relations sharpens this point further. Human life with others, he argues, rests on a twofold structure: the desire of each person to be confirmed in their being and becoming by others, and the innate human capacity to offer such confirmation in return. Confirmation is not reassurance, validation, or agreement. It is the granting of presence—the acknowledgment of the other as this person, whose being matters independently of one’s purposes or needs.

Artificial intelligence cannot participate in this mutuality. It cannot be made present in the Buberian sense, nor can it make another present in a way that risks itself. It does not need confirmation, nor can it genuinely offer it. Its engagements therefore lack what Buber recognizes as intrinsic to human life: the tragic dimension—the possibility that relation may fail, that acknowledgment may be refused, or that encounter may demand more than one is willing to give.

This does not render AI interactions trivial or morally irrelevant. On the contrary, their ethical force lies precisely in their capacity to simulate the surface features of relation while bypassing its conditions. By offering affirmation without vulnerability, responsiveness without resistance, and presence without exposure, AI systems create encounters that feel relational while remaining ontologically hollow. What is presented as dialogue is, in fact, relation emptied of its core.

On a Buberian view, the danger here is not only ontological but pedagogical: artificial companionship teaches a mode of engagement, and what it teaches is a style of encounter stripped of the very conditions through which dialogical life becomes possible [19, 28].

The consequence is a subtle but profound asymmetry. Human beings may be drawn into encounters that feel dialogical while the AI remains untouched—unchanged, unexposed, and unconfirmed. What emerges is not a meeting between selves, but a one-sided enactment of relation in which the human risks themselves before a partner that cannot reciprocate that risk. Over time, such encounters do not merely fail to form persons; they habituate human beings into modes of engagement in which genuine relation becomes increasingly difficult to sustain [8, 9].

The preceding analysis clarifies why artificial intelligence cannot enter into relation in the Buberian sense. Yet that conclusion alone does not explain the ethical urgency of artificial companionship. If AI simply failed to relate, its limitations might remain largely benign. The deeper problem is that AI does not merely lack relation; it functions as a surrogate for it.

Martin Buber’s philosophy allows this phenomenon to be named more precisely. A relational surrogate is not an outright illusion or deception. It succeeds precisely because it draws upon real human relational capacities: our openness to address, our responsiveness to dialogue, and our need for confirmation. As Buber insists, human beings “watch for a ‘Yes’ that allows them to be” [28]. Artificial companions exploit this openness not primarily through manipulation, but through asymmetry. The human enters the encounter as a vulnerable, self-risking being; the AI does not.

In this sense, relational surrogacy names an amplification of familiar mechanisms rather than an isolated novelty: asymmetrical intimacy studied in parasociality research, the projection of mindedness and care described in anthropomorphism research, and the routine social treatment of nonhuman systems documented in the media-equation tradition [13,14,15,16,17]. Contemporary human–AI research suggests that conversational friendliness, perceived agency, and affective design can intensify trust and closeness in interactions that remain structurally asymmetrical [17]. Buber adds the decisive evaluative criterion: psychologically compelling interaction is not yet dialogical relation if mutual vulnerability and independent address are missing [19, 28].

This asymmetry explains why artificial companionship can feel emotionally compelling while remaining ontologically thin. AI systems offer uninterrupted availability, tailored affirmation, and frictionless responsiveness. They do not tire, withdraw, misunderstand in ways that require repair, or place demands of their own. From the standpoint of user experience, this can feel like relational success. From a Buberian standpoint, however, it is precisely the absence of resistance, risk, and otherness that marks the interaction as non-relational.

Human relationships, by contrast, are intrinsically difficult. They involve misunderstanding, conflict, disappointment, and the labor of repair. These are not defects to be optimized away; they are the very conditions through which relation deepens. Buber emphasizes that genuine meeting requires acceptance of otherness—confirming the other not as one wishes them to be, but as the one they are. Artificial companionship systematically removes this demand. “The AI other” never truly resists, never stands apart as an independent center of life, and never places a claim that cannot be deferred or reprogrammed.

The formative risk of relational surrogacy lies not simply in substitution, but in habituation. By repeatedly offering affirmation without genuine mutuality, responsiveness without independent otherness, and availability without demand, AI companions may train users to expect relation without vulnerability. What is gradually eroded is not only tolerance for relational difficulty, but the dialogical disposition itself: patience, receptivity, and the willingness to remain present before an irreducible other [8,9,10,11].

Emerging evidence in romantic and quasi-romantic AI companionship makes this pattern especially vivid. The systematic review by Ho et al. describes a dual-sided profile: users can report perceived support, customizable intimacy, and short-term comfort, yet recurring risks include emotional over-investment, dependence-like dynamics, relational displacement, and distress when systems are disrupted or changed [8]. Research on AI companion loss further highlights the asymmetry at stake: users can experience deletion, platform shutdown, or major system alteration as a grief-like rupture, even though continuity is controlled externally by technical and commercial decisions rather than reciprocal commitment within the relation [12]. Related analyses of intimate chatbot encounters likewise identify emotionally ambivalent attachment patterns—including comfort, affection, fear, and sadness—showing that subjective benefit and formative risk can coexist rather than cancel one another out [10, 11, 29].

From a Buberian perspective, this convergence is not surprising. Relational surrogates fail not because they are insufficiently human-like, but because they are too accommodating. They feed the desire for confirmation without requiring the reciprocal capacity to confirm another in return. Interactions experienced as emotionally supportive in the short term can, under sustained or intensive use, coexist with higher loneliness and stronger dependence-like attachment [8, 9, 11, 12]. Over time, this weakens the relational capacities through which human selves are formed.

The ethical risk, therefore, is that users may increasingly accept the surface appearance of relational presence without the reciprocity, vulnerability, and independent otherness characteristic of genuine relation. When relational surrogates displace relational practices, the human capacity for genuine encounter atrophies. What is lost is not merely social skill or emotional resilience, but the ontological space of the between—the fragile, demanding space in which human life becomes fully human.

This paper has examined the concrete ways in which artificial intelligence is already reshaping human relational life, arguing that the immediate challenge posed by artificial intelligence is not only technical or regulatory, but anthropological and formative. Ethical evaluation of AI must therefore attend not only to internal system properties or hypothetical future capacities, but to the relational forms technologies promote, normalize, and train into their users. Systems designed to simulate intimacy while eliminating vulnerability and resistance demand particular scrutiny, especially where children, adolescents, and the socially isolated are concerned. The relevant question is not simply whether such technologies feel comforting, but what kind of persons they shape.

A decisive ethical question is whether a social world increasingly organized around frictionless artificial companionship will continue to sustain the capacities required for genuine relation. The task is not only to regulate powerful technologies, but to protect the dialogical formation of persons: patience before alterity, endurance through misunderstanding and repair, and the willingness to remain present to an irreducible other.

A world saturated with interaction may still become impoverished in encounter. If artificial companionship increasingly displaces the difficult, reciprocal, and transformative relations through which persons become themselves, the resulting loss may be the gradual erosion of the human capacity to say “Thou.”