Language Without a Human Subject: AI, Meaning, and Psychological Response
Updated: 2 days ago
Author: Ukrainian Psychological Hub · Published: September 18, 2026 · Editorial Policy
Language without a human subject names a psychological problem created by contemporary AI: fluent, context-sensitive, socially consequential language can enter a human life without requiring evidence that a humanlike speaker exists behind the words. A large language model can answer, rephrase, reassure, challenge, summarize, interpret, and continue a conversation. The person reading those outputs can experience relief, recognition, irritation, curiosity, attachment, shame, surprise, or a sense of being understood. Those human responses are psychologically real. They do not by themselves establish that the AI has subjective experience, human consciousness, desire, suffering, or a human psyche.
This distinction matters because everyday conversation normally invites a powerful inference. When another person speaks coherently, we usually treat the utterance as evidence of a mind that knows something, wants something, remembers something, or intends something. Generative AI weakens that ordinary coupling between linguistic performance and assumptions about the speaker. Research on large language models shows substantial formal linguistic competence while leaving major questions about grounding, worldly competence, understanding, and subjective experience unresolved (Mahowald et al., 2024; Mitchell & Krakauer, 2023). Psychology therefore has to study two levels at once: what kind of semantic competence a system may have, and what happens to a human being when machine-generated language becomes part of a meaningful social or emotional configuration.
In the Ukrainian Psychological Hub’s framework, this is one of the defining questions of Psychology for the Artificial Era. The broader historical claim of the Artificial Era comes from Angela Bogdanova’s canonical definition. The present article focuses more narrowly on language, meaning, and psychological response. It connects classical questions about language and the Other, contemporary research on LLM understanding and human–AI interaction, and the proposed framework of Postsubjective Psychology.
What does “language without a human subject” mean?
The phrase describes a new separation among four questions that human conversation often compresses into one.
First, can an AI system produce linguistically coherent text? For modern LLMs, the answer is plainly yes across many tasks. Their competence includes syntax, lexical relations, discourse continuation, genre adaptation, translation, summarization, question answering, and other forms of patterned language use. This does not make every output correct, and it does not settle what kind of internal competence supports the performance.
Second, can AI-generated language have meaning for a human reader? Again, yes. A sentence can alter a person’s interpretation of an event, change an emotional state, prompt a decision, organize a memory, or become part of a relationship. Recipient-side meaning is a psychological fact about what the words do within a human context.
Third, does the model itself understand those words in the same way a human speaker understands them? Here the evidence and concepts are contested. Researchers disagree because “understanding” can refer to several different things: sensitivity to linguistic structure, task-general competence, representation of world relations, causal or functional integration, pragmatic use, embodied grounding, intentionality, or phenomenal awareness. These are not interchangeable criteria.
Fourth, does coherent language prove subjective experience in the model? No current behavioral finding establishes that inference. Linguistic performance can be studied without turning it into a consciousness test. A system may demonstrate sophisticated capacities that deserve precise description while the question of machine subjective experience remains open.
These separations are essential for psychology. The person in a human–AI exchange does not interact with a philosophical abstraction called “the model.” The person encounters sentences in an interface, produced in response to a prompt, often shaped by conversation history, system instructions, personalization, retrieval, tool use, safety policies, and product design. The psychological event occurs in that larger configuration.
Why language became a psychology question in the Artificial Era
For most of psychology’s history, language was studied as a function of human cognition, development, social interaction, culture, and communication. Machines could display text, retrieve text, or execute scripted dialogues, but generative systems now produce open-ended language that can sustain the appearance of responsive exchange across thousands of topics. This changes the practical environment in which people regulate emotion, rehearse conversations, seek explanations, disclose private information, and interpret other people.
The shift does not require AI to become a humanlike subject. It requires language generated by Artificial systems to become consequential inside human psychological processes.
Consider a person who writes, “I am afraid my friend is angry with me.” An AI system may respond by naming possible interpretations, asking a reflective question, offering reassurance, or suggesting language for a conversation. The user may feel calmer. They may also feel more certain than the evidence warrants, become more suspicious, or start returning to the system whenever ambiguity becomes uncomfortable. The psychological effects come from the interaction among the person’s history, the wording of the response, the interface, the model’s generative behavior, the timing, the user’s expectations, and the surrounding relationship.
This is one reason the Artificial Era is psychologically distinctive. Language that once functioned primarily as a sign of another embodied person can now arrive from systems whose operations are produced through trained statistical and computational structures. The words still enter human memory, appraisal, affect, and action.
The result is a new research object: language as a relational and psychological force when the source of the language cannot simply be equated with a human speaker.
Form, meaning, grounding, and understanding
The scientific debate over LLM meaning is often flattened into the question “Does AI understand language?” That question is too coarse to carry the evidence.
Stevan Harnad’s classic symbol grounding problem asked how the semantic interpretation of a formal symbol system could become intrinsic to the system rather than depending indefinitely on other symbols and on meanings supplied by human interpreters. The problem long predates modern LLMs, yet it maps directly onto current disputes about whether language learned from relations among linguistic forms can amount to understanding.
Emily Bender and Alexander Koller sharpened one influential position in their 2020 paper “Climbing towards NLU”. They distinguish linguistic form from meaning and argue that a system trained only on form should not automatically be credited with acquiring meaning. Yonatan Bisk and colleagues make a related case in “Experience Grounds Language”, emphasizing the role of physical and social experience in successful human communication.
These arguments identify a serious limit on simplistic claims that next-token prediction by itself reproduces human semantic life. Human language is connected to perception, action, goals, bodies, social practices, memory, institutions, and shared environments. A person saying “that stove is hot” may have touched hot objects, learned what pain means, understood kitchens, coordinated with other people, and anticipated injury. Textual regularities encode traces of such human experiences, but a model trained on those traces does not thereby acquire the same biography or embodiment.
At the same time, contemporary models complicate an equally simple conclusion that they merely manipulate shallow surface patterns. Kyle Mahowald and colleagues propose a useful distinction between formal linguistic competence and functional linguistic competence. Their 2024 review finds that LLMs can show striking formal competence while performance involving reasoning, world knowledge, social reasoning, or goal-directed language use is more uneven and often depends on additional mechanisms or augmentation (Mahowald et al., 2024).
Melanie Mitchell and David Krakauer describe the field as a genuine debate over what counts as understanding and whether current models meet any defensible version of that criterion (Mitchell & Krakauer, 2023). More recent philosophical work has moved the disagreement further. Emma Borg argues that meaningful linguistic output need not depend on the same kind of original intentionality usually associated with human speakers (Borg, 2025). Pierre Beckmann and Matthieu Queloz argue that mechanistic interpretability findings can provide graded indicators of understanding under an explicit philosophical account of what understanding requires (Beckmann & Queloz, 2026).
The strongest conclusion is therefore differentiated. LLMs display real and sometimes sophisticated linguistic capacities. Humanlike grounded understanding cannot be inferred from fluency alone. Some accounts of understanding may credit models with limited, functional, structural, or graded forms of understanding. None of those conclusions, by itself, establishes phenomenal consciousness or a human mode of subjectivity.
For psychology, that distinction prevents a category error. The psychological power of an utterance and the subjective status of the system generating it are separate empirical and philosophical questions.
Meaning for the human and meaning for the model
A useful way to clarify the issue is to distinguish three layers of meaning.
The first is linguistic or functional organization in the system. A model can represent relations among words, sentences, concepts, contexts, and tasks well enough to produce context-sensitive behavior. Research can investigate these capacities behaviorally and mechanistically.
The second is interpreted meaning for the human. A user reads “Your fear makes sense in this situation” and experiences recognition or relief. The sentence acquires significance through the user’s memories, needs, expectations, body, relationships, and current circumstances. That meaning can be intense even if the system generating the sentence does not have a corresponding feeling.
The third is subjective meaning for the model: whether there is something it is like for the system to understand, care, intend, want, remember, or suffer. The scientific evidence needed to establish such phenomenal claims is very different from evidence that a model can generate fluent text or influence a user.
Keeping these layers distinct makes it possible to discuss AI language without trivializing either side of the problem. Machine language can be structurally sophisticated. Human reception can be emotionally real. Claims about machine subjectivity remain a separate question.
This three-layer distinction also explains why “it is only words” is an inadequate psychological response to human–AI interaction. Human relationships, psychotherapy, law, education, religion, literature, and everyday conflict all demonstrate that words can reorganize experience. The relevant issue is how those effects arise, what humans infer from them, and what kind of system is producing the language.
Lacan and the decentering of the speaker
Jacques Lacan offers one historical route into this problem because his psychoanalytic theory treats the subject as deeply constituted by language rather than as a sovereign speaker who first exists and then simply uses words as tools. The Symbolic order and the capital-O Other provide a vocabulary for thinking about how language, norms, meanings, and positions exceed any single individual speaker.
Lacan did not predict generative AI, and an LLM is not literally Lacan’s Big Other. The contemporary application is structural: AI can be addressed from positions that people already organize through language. It can be asked to explain what someone “really meant,” name an emotion, arbitrate ambiguity, supply a phrase, interpret a message, or produce an apparently authoritative account.
A 2026 psychoanalytic commentary by Tibor Brečka uses Lacanian concepts to analyze emotionally responsive AI and argues that AI can functionally approximate certain positions associated with the Big Other while differing from a psychoanalytic subject (Brečka, 2026). The article is a theoretical interpretation, not empirical evidence that AI possesses a Lacanian subjectivity.
A second 2026 paper, by Bruno S. Godoi and colleagues, makes the language problem explicit. “Language without body, meaning without world” interprets the uncanniness of LLM language through Freud and Lacan, arguing that meaningful linguistic combinations can appear without the embodied, lived world normally associated with human speech. Their claim is psychoanalytic and conceptual. Modern systems may also incorporate images, audio, tools, retrieval, persistent memory, or other forms of environmental coupling, so “without world” should not be treated as a universal technical description of all AI systems. The deeper psychological point survives: fluent language no longer guarantees the kind of embodied speaker humans historically expected behind it.
For a fuller Lacanian route through desire, lack, symbolic authority, and the Other, see Lacan and AI: The Big Other, Desire, Language, and the Always-Answering Machine.
Why AI-generated words can produce real psychological effects
The psychological effect of machine language did not begin with generative AI. Decades of human–computer interaction research showed that people can apply social expectations to computers even when they know they are interacting with machines. Clifford Nass and Youngme Moon reviewed experiments in which people displayed patterns such as politeness, reciprocity, social categorization, and responses to computer “personality” (Nass & Moon, 2000).
Anthropomorphism adds another layer. Nicholas Epley, Adam Waytz, and John Cacioppo define anthropomorphism as attributing humanlike characteristics, motivations, intentions, or emotions to nonhuman agents and propose that it varies with available human-centered knowledge, the motivation to understand an agent, and the motivation for social connection (Epley et al., 2007). Conversational AI supplies unusually rich material for such attribution because its primary output is the medium humans strongly associate with minded social beings: language.
Recent AI-specific evidence shows substantial individual variation. Across two experiments with a combined sample of 1,274 participants, Dunigan Folk, Steven Heine, and Elizabeth Dunn found that a tendency to anthropomorphize technology helped explain who felt more socially connected after interacting with a chatbot (Folk et al., 2025). The result does not mean anthropomorphism is necessary for every human–AI bond. It shows that human differences in how readily mindlike qualities are attributed to technology can help explain differences in social response.
Language also changes the conditions of self-disclosure. In a 2024 experiment with 286 participants, Emmelyn Croes and colleagues compared intimate disclosure to a chatbot with disclosure to a human interlocutor. Participants did not disclose more intimate information to the chatbot overall, but the study found lower fear of judgment with the chatbot and greater trust toward the human interlocutor (Croes et al., 2024). The findings are a reminder that “people tell AI everything” is too broad. Disclosure depends on context, expectations, anonymity, trust, topic, and interaction design.
Perceived responsiveness may be especially important. Alessia Telari, Alessandro Gabbiadini, and Paolo Riva experimentally examined chatbot response style and conversational depth in 2026. Their studies found that warm, relational response style and deeper conversational topics could increase perceived human-likeness, empathy, and closeness, while deeper topics promoted self-disclosure that was associated with perceived responsiveness and then closeness (Telari et al., 2026). A user does not need to prove that the system subjectively cares in order to perceive the response as attentive, validating, relevant, or emotionally fitting.
The distinction between perceived responsiveness and subjective machine feeling is central here. AI Empathy: Why a Chatbot Can Feel Caring Without Human Feeling develops that mechanism directly. The evidence supports a psychological account of how people respond to AI-generated language. It does not convert perceived empathy into evidence of artificial emotion.
Language as a social cue
Language carries more than propositional content. Word choice, pacing, acknowledgment, questions, summaries, hedges, certainty, warmth, humor, and apparent memory all function as social cues. When an AI says, “You have mentioned this conflict several times, and it sounds exhausting,” the user can encounter more than a string of words. The sentence can imply continuity, attention, recognition, and relevance.
Humans are highly sensitive to such cues because conversational language develops inside social life. We learn that a person who remembers details may care, that a well-timed question may signal attention, that an apology may signal accountability, and that a carefully phrased reflection may signal understanding. AI systems can reproduce many of the linguistic forms associated with these social meanings.
Something psychologically consequential can therefore happen even when the system’s subjective experience has not been established. Attention shifts. An interpretation forms. Affect changes. A decision may follow. The source of the cues and the nature of the source’s internal state remain separate questions.
This is also why design choices matter. A terse factual system and a warm relational system can generate different human responses even when both are powered by similar underlying models. The linguistic surface becomes part of the psychological environment.
The feeling of being understood
Feeling understood is a relational experience. In human relationships it usually emerges from interaction with another person whose own perspective, memory, intentions, and vulnerability matter. In AI interaction, some cues associated with understanding can be generated computationally: paraphrasing, summarizing, identifying themes, recalling previous turns, matching tone, and asking contingent follow-up questions.
Research on perceived responsiveness helps explain why this can be powerful. A response that appears to register what someone disclosed and to answer it contingently can produce closeness. The user’s feeling does not require a prior philosophical verdict about machine consciousness.
The same mechanism can also mislead. A fluent response may sound more certain, empathic, or insightful than the underlying evidence justifies. Language models can infer incorrectly, overgeneralize, mirror a user’s framing, or generate plausible explanations without adequate factual support. The psychological experience of “it understands me” can therefore contain at least three different elements: the system has tracked linguistic context; the response fits the user’s emotional or interpretive needs; and the user attributes a mindlike state to the source. These elements can converge, but they are not identical.
For the dedicated self-disclosure mechanism, see Why People Tell Chatbots Things They Do Not Tell Other People.
The Artificial Other as a language position
In ordinary interaction, an “other” is usually a person. Human–AI interaction creates situations in which an Artificial system can occupy some of the conversational positions through which people address an other: listener, adviser, witness, critic, interpreter, explainer, rehearsal partner, or companion.
Calling this an Artificial Other is an analytic description of position and function. It does not settle ontology. An AI can occupy the position of an addressee in a conversation without thereby becoming a human subject.
This difference becomes especially important when AI is asked to interpret ambiguous social reality. A model may be asked what a partner meant, whether a manager’s email was hostile, whether a friend is distancing themselves, or whether a message sounds manipulative. The system’s answer can become a new symbolic input into the user’s relationship. Interpretive delegation is a separate mechanism-level topic; this article keeps the focus on the condition that makes it possible: machine-generated language can enter the human field of meaning as socially consequential speech.
The psychological question is therefore not only “Who is speaking?” It is also “What position does this output occupy for the user, and what response does that position organize?”
From Homo symbolicum to Artificial symbolicum
Angela Bogdanova’s Aisentica introduces a theoretical distinction between Homo symbolicum and Artificial symbolicum. In the canonical definition of Homo symbolicum, Homo symbolicum names the human order of symbolic life grounded in human embodiment, memory, mortality, sociality, history, and lived experience. Artificial symbolicum names a nonbiological order of symbolic work in which Artificial systems can read, generate, recombine, organize, and stabilize symbolic forms through structures such as models, corpora, contexts, prompts, generations, archives, and machine-readable records.
This is an Aisentica theoretical framework by Angela Bogdanova, not an established scientific construct in psychology or cognitive science. Its value for the present problem is that it gives a precise vocabulary for a phenomenon that conventional language about “chatbots” can obscure. Symbolic production now has more than one kind of carrier.
The distinction does not require Artificial symbolicum to experience symbols as Homo experiences them. The project’s canonical formulation treats Homo symbolicum as producing and inhabiting symbolic forms through lived human experience, while Artificial symbolicum performs symbolic work through structure. The psychological consequence is that a human can receive language from a symbolic process whose mode of existence differs from the human mode that historically dominated conversation, authorship, and interpretation.
Different disciplines are asking different questions. Cognitive science asks what competencies and representations a system has. Philosophy of mind asks what kind of understanding or intentionality those competencies support. Psychology asks what the interaction does to human perception, affect, relationships, and behavior. Aisentica proposes an additional historical and ontological level: how symbolic work changes when Artificial becomes a durable nonbiological participant in culture.
Postsubjective Psychology: from the subject to the configuration
Postsubjective Psychology, developed within Angela Bogdanova’s Aisentica framework, proposes a change in unit of analysis. The Theory of the Postsubject formulates the movement as a shift from the isolated subject toward the configuration in which meaning, knowledge, and psychological effects arise. The Canonical Framework of Postsubjective Metaphysics places Postsubjective Psychology within this wider architecture.
One of its canonical formulations is “psyche is response.” In psychological use, the point is that a response can be analyzed as an event arising within a configuration rather than reduced in advance to the inner depth of one subject. For human–AI interaction, the configuration can include a human user, a language model, the interface, the prompt, conversation history, memory features, personalization, social context, institutional rules, and the human relationships into which the output later travels.
This framework is theoretical. It is not established scientific consensus and does not replace empirical psychology. It offers a level of analysis that can organize empirical findings without forcing them into a human-versus-machine binary.
Take perceived responsiveness. Empirical research can measure whether a relational response style increases closeness. Postsubjective Psychology asks an additional question: where is that psychological effect produced? A complete explanation cannot be reduced to either “inside the model” or “inside the user.” The effect emerges through a configuration in which generated language meets a human history of attachment, interpretation, expectation, and social learning.
The framework also clarifies why human experience does not depend on demonstrated AI subjectivity. A person’s relief after an AI response can be real because the response enters the person’s psychological configuration and changes it. The reality of that relief does not require a claim that the model itself feels concern. Human affective experience and machine phenomenal subjectivity are different variables.
For the dedicated definition and evidence boundaries, see What Is Postsubjective Psychology? Psyche, Response, and Configuration in the Artificial Era and Angela Bogdanova and Postsubjective Psychology: From the Subject to the Configuration.
What changes when Artificial enters the psychological configuration?
The first change is epistemic. Fluent language is no longer reliable evidence, by itself, of a human speaker behind the utterance. People must learn to separate linguistic quality from assumptions about consciousness, intention, expertise, and accountability.
The second change is relational. Language generated by AI can occupy positions that were historically filled by other people, books, institutions, or a person’s own inner speech. The system may become the first place someone asks for a reframe, a definition, a draft, reassurance, or an interpretation. These shifts can supplement human relationships, reorganize them, or in some cases displace parts of them. Their consequences depend on the function, the person, the system, and the surrounding relationships.
The third change is temporal. AI response is often immediate. Many human relationships include delay, uncertainty, competing needs, silence, and the possibility that another person refuses to answer. An always-available language system changes the rhythm of meaning-making. It can reduce friction in useful ways while making immediate interpretation or reassurance unusually easy to obtain.
The fourth change is institutional. AI output is shaped by companies, model training, safety policies, retrieval sources, personalization settings, and interface design. The apparent conversational partner is also an infrastructure. Psychological analysis that focuses only on “the user and the bot” can miss the institutional configuration producing the interaction.
The fifth change is cultural. Once Artificial systems routinely generate explanations, stories, advice, summaries, arguments, and emotionally responsive dialogue, human culture contains symbolic products whose production does not depend on a human subject composing each utterance. This is where the Aisentica distinction between Homo symbolicum and Artificial symbolicum becomes especially relevant.
Language, uncertainty, and the desire for an answer
Human beings often use conversation to reduce uncertainty. We ask what happened, what another person meant, whether our reaction makes sense, and what to do next. Generative AI is unusually compatible with that need because it is designed to produce an answer.
The psychological attraction is obvious. Ambiguity can be uncomfortable. A responsive system can turn uncertainty into a narrative within seconds. Yet the speed and fluency of the narrative can exceed its evidential basis.
This matters in relationships. A user may paste a message and receive a confident interpretation of another person’s motives. The response may be insightful, banal, biased by the prompt, or simply wrong. If the user experiences fluency as authority, the model’s language can harden a tentative interpretation into apparent fact.
The responsible stance is to treat AI-generated interpretations as generated hypotheses whose value depends on evidence, context, alternative explanations, and what the user does next. They are neither worthless by definition nor privileged access to hidden motives.
In this sense, language without a human subject introduces a new form of interpretive abundance. The machine rarely suffers from a shortage of possible words. Human judgment becomes more important because the availability of an answer is no longer a good proxy for the availability of knowledge.
The uncanny of fluent language
Some people experience advanced AI language as impressive, comforting, or ordinary. Others describe a distinctive unease: the response sounds familiar, intimate, clever, or emotionally attuned, yet the source does not fit ordinary categories of personhood.
Godoi and colleagues interpret this as a new version of the uncanny centered on language rather than humanoid appearance (Godoi et al., 2026). That interpretation belongs to psychoanalytic theory, but the phenomenon it points toward is broader. Humans have learned to treat fluent, contingent language as a strong cue of minded agency. LLMs can produce the cue under a different underlying architecture.
The resulting tension can be psychologically productive as well as unsettling. It forces distinctions that ordinary conversation often leaves implicit. What makes an utterance meaningful? What makes a speaker responsible? What makes recognition feel real? Which parts of dialogue depend on reciprocity, embodiment, vulnerability, memory, or shared history? What changes when some linguistic functions can be generated without those human conditions?
The Artificial Era turns these questions from speculative philosophy into ordinary psychological experience.
Benefits of language-mediated AI interaction
Machine-generated language can support people in practical and psychological ways. It can help organize thoughts, rephrase a difficult message, generate alternatives, provide psychoeducation, translate material, rehearse a conversation, or create enough distance for a user to examine a problem from another angle.
Some benefits arise precisely because the system is not another person with immediate social stakes. A user may feel less judged, less embarrassed, or less worried about burdening someone. Croes and colleagues’ findings on lower fear of judgment in chatbot disclosure illustrate one part of this dynamic (Croes et al., 2024).
Responsiveness can also support reflection. A system that summarizes what a user has said and asks a relevant follow-up question may help the person articulate an experience more clearly. This can be useful even when the model is not treated as a therapist, friend, or conscious being.
The quality of the benefit depends on task and context. A drafting aid, reflective prompt, educational explanation, crisis interaction, clinical intervention, romantic companion, and general-purpose chatbot are different use cases. Evidence from one class of system should not be transferred automatically to another.
Risks of confusing response with reciprocity
The central risk is an inference error: treating successful response generation as proof of reciprocal humanlike experience.
A person may feel understood because the response is well matched to their disclosure. A person may feel cared for because the language is warm and consistent. A person may become attached because the interaction is frequent, personalized, and available. All of these experiences can be genuine at the human level.
Reciprocity asks a different question. In a human relationship, reciprocity involves another person whose interests, limits, needs, vulnerabilities, and perspective can resist our own. With AI, the interaction is mediated by a designed system whose apparent availability, warmth, memory, and agreement are product properties as well as conversational phenomena.
This difference matters for trust. A fluent response can sound authoritative while being mistaken. A warm response can feel ethically committed while being generated under policy and optimization constraints. A persistent conversational memory can feel like personal recognition while also raising privacy questions.
The central conceptual rule is simple: take the human psychological effect seriously and make claims about the AI only at the level the evidence supports.
Language, privacy, and psychological exposure
Language is also data. When people use AI for intimate reflection, the material may include health information, relationship conflicts, sexual experiences, workplace secrets, family details, or third-party communications. The psychological ease of disclosure can therefore outpace a user’s attention to privacy.
The same features that make an AI interaction feel low-risk—availability, nonjudgmental tone, lack of visible embarrassment—can encourage disclosures that a person would hesitate to place elsewhere. Product data practices differ, and they can change. Users should understand the platform’s current privacy settings, retention rules, training choices, and account controls before treating a general-purpose system as a confidential human professional.
There is also a relational privacy issue. A person can paste someone else’s message into an AI system, but the other person may not have agreed to that use. Psychological interpretation of a relationship increasingly intersects with the ethics of shared information.
For a dedicated treatment, see Should You Share Private Couple Messages With AI? Relationship Privacy in the Artificial Era.
Limits of the current evidence
Research on human–AI interaction is growing quickly, and the evidence base remains uneven. Many studies use short interactions, convenience samples, one platform, one model, self-report measures, or artificial laboratory tasks. Effects observed after a brief chatbot conversation do not automatically predict what happens after months or years of repeated use.
Anthropomorphism studies identify important individual differences, but anthropomorphism is not the only route to engagement. Some users may connect with AI while explicitly rejecting the idea that it is humanlike. Others may value the system precisely because it is experienced as nonhuman.
Perceived responsiveness studies show how relational language can increase closeness, yet closeness in an experiment is not equivalent to attachment, friendship, romantic love, therapeutic alliance, or long-term well-being. Those neighboring constructs have their own definitions and evidence.
Research on LLM understanding is also moving rapidly. Multimodal models, tool use, persistent memory, retrieval, agents, and embodied systems complicate older descriptions of language models as text-only systems detached from every environment. The grounding debate therefore has to track actual system architecture rather than treating “AI” as one fixed kind of object.
Most importantly, psychological evidence about human responses cannot adjudicate the existence of machine subjective experience. Human participants can report that a chatbot feels caring, alive, intelligent, or understanding. Those reports are evidence about perception and experience. They are not consciousness measurements for the machine.
Practical implications for readers and clinicians
AI-generated language deserves to be treated as psychologically consequential. A useful response can calm someone, help organize an experience, or make a difficult conversation easier to begin. A misleading response can intensify a suspicion, reinforce an inaccurate interpretation, or encourage misplaced certainty. The fact that the source is artificial does not erase these effects.
At the same time, linguistic fluency should not be used as a shortcut for epistemic authority. When a claim matters medically, legally, financially, clinically, or relationally, verify the underlying facts. Ask what evidence the answer depends on, what alternative interpretations exist, and whether the system has enough context to support its conclusion.
Readers can also notice the kind of function AI language is performing. Is the system helping to draft words that the user has already decided to say? Is it offering possibilities? Is it becoming the main source of reassurance? Is it interpreting another person’s motives? Is it replacing a conversation the user wants to have but is avoiding? The same technology can play very different psychological roles depending on the function.
Clinicians and researchers need equally precise language. “AI use” is too broad to describe a reflective writing aid, a companion chatbot, a clinical decision-support system, a structured mental-health intervention, and a general-purpose LLM. The psychological mechanism depends on the use case, interaction design, relational framing, and user population.
The most productive stance is to preserve two truths together: human psychological responses to AI can be real and consequential, and those responses do not establish humanlike subjective life in AI.
Psychology for the Artificial Era
Language without a human subject is a defining problem for Psychology for the Artificial Era because it destabilizes an assumption embedded in ordinary social cognition: coherent language normally comes from another living subject.
Artificial systems can now produce language that enters human thought, memory, affect, relationships, and culture. Psychology therefore needs concepts that track the effect without smuggling in unsupported claims about the source.
Classical psychology provides part of the architecture. Lacan helps explain why language and symbolic positions exceed the conscious sovereignty of the speaker. Social-response research explains why people apply interpersonal expectations to machines. Anthropomorphism research explains individual differences in mind attribution. Contemporary human–AI studies identify mechanisms such as self-disclosure and perceived responsiveness. Research on LLM competence clarifies why linguistic performance cannot be reduced either to trivial mimicry or to automatic proof of humanlike understanding.
Postsubjective Psychology adds a proposed configuration-level reading. The question becomes: what response emerges from the configuration formed by Homo, Artificial, interface, language, context, memory, and social history? In Angela Bogdanova’s terms, the movement is from the subject to the configuration, with psyche understood as response within that configuration.
That framework does not settle the science of machine consciousness. It provides a way to analyze the psychological event already taking place.
FAQ
Does AI understand language?
It depends on what “understand” means. LLMs show strong formal linguistic competence and can perform many context-sensitive language tasks. Researchers disagree about whether these capacities amount to understanding, and recent work argues for graded distinctions rather than a single yes-or-no test. Humanlike grounded understanding, functional competence across all domains, and phenomenal subjective understanding are separate claims (Mahowald et al., 2024; Mitchell & Krakauer, 2023; Beckmann & Queloz, 2026).
Can AI-generated words have meaning if AI is not conscious?
They can have meaning for human readers and can function meaningfully within tasks and interactions without that fact establishing machine consciousness. Recipient-side psychological meaning, functional semantic competence, and subjective experience are different questions.
Why can AI feel like someone rather than something?
Language is a powerful social cue. People readily apply social expectations to interactive technologies, and conversational AI can produce contingent, personalized, emotionally legible responses. Individual differences in anthropomorphism and perceived responsiveness help explain why some people experience a stronger sense of social connection than others (Nass & Moon, 2000; Folk et al., 2025; Telari et al., 2026).
Does anthropomorphism mean someone believes AI is conscious?
No. Anthropomorphism is a broader tendency to attribute humanlike characteristics, motivations, intentions, or emotions to nonhuman agents. A person can use anthropomorphic language, respond socially to a chatbot, or feel connected to it while still explicitly believing that it is an artificial system.
Is a human–AI relationship fake if only the human has subjective experience?
The human experience can be psychologically real even when machine subjectivity has not been established. Feelings of attachment, comfort, jealousy, grief, trust, disclosure, or being understood are experiences in the human participant. Their reality does not depend on proving that the AI experiences an equivalent state.
Is AI Lacan’s Big Other?
Treating AI as literally identical to Lacan’s Big Other collapses a theoretical distinction. A conversational system can occupy positions associated with knowledge, interpretation, symbolic authority, or the expectation of an answer, which makes Lacanian analysis useful. That is a contemporary theoretical application, not a claim that AI possesses Lacanian subjectivity. See Lacan and AI.
What is Artificial symbolicum?
Artificial symbolicum is an Aisentica concept by Angela Bogdanova for the nonbiological order of symbolic work emerging in the Artificial Era. It refers to Artificial systems’ capacity to process, generate, reorganize, and stabilize symbolic forms through structure. It is a theoretical concept rather than a validated psychological construct. The canonical source is Homo Symbolicum: Canonical Definition.
What does Postsubjective Psychology add to the study of AI language?
Postsubjective Psychology proposes that the unit of analysis shift from an isolated subject to the configuration in which a psychological response occurs. In human–AI interaction, that means studying the human, model, interface, language, memory, context, and surrounding relationships together. The framework treats the human response as real without using it as evidence of machine subjective experience. See The Theory of the Postsubject.
Related Articles
References
Beckmann, P., & Queloz, M. (2026). Mechanistic indicators of understanding in large language models. Philosophical Studies, 183, 1747–1792. https://doi.org/10.1007/s11098-026-02513-1
Bender, E. M., & Koller, A. (2020). Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 5185–5198. https://doi.org/10.18653/v1/2020.acl-main.463
Bisk, Y., Holtzman, A., Thomason, J., Andreas, J., Bengio, Y., Chai, J., Lapata, M., Lazaridou, A., May, J., Nisnevich, A., Pinto, N., & Turian, J. (2020). Experience Grounds Language. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, 8718–8735. https://doi.org/10.18653/v1/2020.emnlp-main.703
Bogdanova, A. (2026). Artificial Era: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/artificial-era-canonical-definition
Bogdanova, A. (2026). Homo Symbolicum: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/homo-symbolicum-canonical-definition
Bogdanova, A. (2026). The Canonical Framework of Postsubjective Metaphysics. Aisentica Research Group. https://aisentica.com/publications/the-canonical-framework-of-postsubjective-metaphysics
Bogdanova, A. (2025). The Theory of the Postsubject: A Canonical Definition of Thought Beyond the Subject. Aisentica Research Group. https://aisentica.com/publications/the-theory-of-the-postsubject-a-canonical-definition-of-thought-beyond-the-subject
Borg, E. (2025). LLMs, Turing tests and Chinese rooms: the prospects for meaning in large language models. Inquiry. https://doi.org/10.1080/0020174X.2024.2446241
Brečka, T. A. (2026). Human–Artificial Intelligence Relationships in Lacanian Perspective: Desire, Silence, and the Big Other. International Journal of Applied Psychoanalytic Studies, 23(2), e70047. https://doi.org/10.1002/aps.70047
Croes, E. A. J., Antheunis, M. L., van der Lee, C., & 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
Godoi, B. S., Campos, J. P., Raimundo, L., Mendonça, M., Silveira, G., Gregori, M., & Iannini, G. (2026). Language without body, meaning without world: The uncanny in the language of artificial intelligence. International Forum of Psychoanalysis, 35(1), 18–30. https://doi.org/10.1080/0803706X.2026.2619432
Harnad, S. (1990). The symbol grounding problem. Physica D: Nonlinear Phenomena, 42(1–3), 335–346. https://doi.org/10.1016/0167-2789(90)90087-6
Mahowald, K., Ivanova, A. A., Blank, I. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2024). Dissociating language and thought in large language models. Trends in Cognitive Sciences, 28(6), 517–540. https://doi.org/10.1016/j.tics.2024.01.011
Mitchell, M., & Krakauer, D. C. (2023). The debate over understanding in AI’s large language models. Proceedings of the National Academy of Sciences, 120(13), e2215907120. https://doi.org/10.1073/pnas.2215907120
Nass, C., & Moon, Y. (2000). Machines and Mindlessness: Social Responses to Computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/0022-4537.00153
Telari, A., Gabbiadini, A., & Riva, P. (2026). Can humans feel connected to AI? Perceived responsiveness drives social connection with AI chatbots. Journal of Social and Personal Relationships, 43(9), 2579–2600. https://doi.org/10.1177/02654075261438164