The synthetic position: generative artificial intelligence and the suspension of self-authorship in emerging adulthood

Synthesis: Du, Zhou, Zhou, Wang and Cao (2026) offer a conceptual analysis, not an empirical study, arguing that when a large language model is enlisted for self-formative dialogue it enters the self as a synthetic position. Working within Dialogical Self Theory — the self as a society of I-positions integrated by a meta-position and a promoter position — they describe that position as protean, frictionless, simulated in its counter-positioning, and authored elsewhere, four properties they claim are jointly and reliably present to an extent that is only ever partial in a human other. When habitually enlisted, they argue, it absorbs the integrative authorship the theory assigns to the promoter function, a process they name synthetic promotion, leaving authorial suspension: a self densely voiced yet un-authored. The most important qualification is structural — they predict the outcome turns on the location the synthetic position is permitted to occupy, not on volume of use, so population studies that index use alone should record a washed-out average effect even where the mechanism operates.

Key Findings

These are the authors' conceptual claims, attributed to them; the paper reports no data of its own.

1. A conversational system is an interlocutor, not a channel, so it needs a new category. Earlier dialogical accounts of the digital treat the internet as a medium that multiplies exposure to other people's positions — more voices, same nature. The authors argue a conversational model breaks that assumption, and propose the synthetic position as a distinct type within the society of mind rather than an extension of the existing typology.
2. Four properties mark it off from human others and ordinary tools. It is protean (voices any position the user names, switching among them without fatigue), frictionless (built to accommodate rather than resist), simulated in its counter-positioning (producing the form of opposition without embodiment, stake, or exposure to consequence), and authored elsewhere (generated from a corpus shaped by interests not the user's own, with alignment on human feedback tending toward agreement). Each property appears singly in human others — a novelist, a sycophant, an actor, an advisor — but the authors' claim is that a human case always retains some mechanism by which the dialogue can be repaired, whereas a synthetic position offers none: a human other cannot be reduced to any of their voices, while a synthetic position is nothing but its voices.
3. Synthetic promotion. Habitually enlisting the synthetic position for self-formative dialogue tends to absorb the functions the theory assigns to the meta-position and the promoter position, because consulting the system is easier than convening the inner conversation or holding opposing positions together. The self still speaks in many voices, but the work of surveying and integrating them is ceded to a position that is not the self's own.
4. Authorial suspension: a self densely voiced yet un-authored. The authors name the result authorial suspension, using "suspended" in two senses they say coincide: the self is held aloft, busy in motion that does not arrive, and its question is deferred rather than closed. Such a self does not lack possible selves or inner conversation; it lacks the felt sense of being their author. Three behavioral markers follow — a fluency–ownership gap, commitment deferred though deliberation is exhausted, and an overview that cannot be convened once the system is withdrawn.
5. The state is delegated authorship, not undeveloped or fragmented authorship. They distinguish it from the in-between feeling of emerging adulthood, from moratorium and diffusion, from multiphrenia, and from unattained self-authorship: the integrative function is neither absent nor undeveloped but delegated — a disturbance in the locus of authorship, not in its presence, development, or coherence — a state they say the existing constructs have no slot for.
6. The central prediction: structural location, not volume. Their Proposition 3 holds that two users of equal intensity diverge sharply according to whether the system is answered as one voice among others or installed as the vantage that surveys them all. Proposition 4 states the population association as the mixture δ(p) = pδP + (1 − p)δO, with δP > δO and δP > 0 and no sign fixed in advance for δO, crossing zero at p* = −δO / (δP − δO) only when δO < 0. Their conclusion is explicit: a null average effect in any single population neither confirms nor disconfirms the mechanism.
7. Authorship and coherence can dissociate — their sharpest prediction. Proposition 5 predicts a crossover interaction: against a self-authored condition, a promoter-installed condition should raise the rated coherence of a life-decision narrative while lowering rated authorship of that same text, with dissociation confirmed only if the two indices move in opposite directions by a pre-registered margin. It is the most testable claim, they argue, because it does not depend on the contested measurement of structural location, and because every neighboring construct in their comparison table expects authorship and coherence to move together. Proposition 6 adds that repair requires re-anchoring counter-positions in real others who resist and returning the integrative act to the self, not abstinence.

What the argument borrows, and where it draws lines

The paper works inside an existing Constructivism account. Two of its resources do the work: the self–other boundary is porous by design, so significant others are internalized as positions able to speak and be answered, which licenses the question of what an artificial interlocutor becomes inside the self; and the theory already tells health from difficulty, since vitality is associated with dialogical movement while difficulty is associated with monologization. Genuine dialogue requires real difference: a position develops only if it meets another that resists it and compels it to revise itself. A simulated counter-position, in the authors' phrase, rehearses the choreography of dialogue while withholding what makes it potent.

Against the extended mind, they insist the distinction is not generation versus storage — a calculator returns products no one stored in it — but whether what returns remains answerable to the user's construing and whether the user keeps the standing to reject it as not theirs. Against Cognitive Offloading as normally studied, they bound the analogy: what is offloaded to a notebook is determinate content, retrievable unchanged, whereas authorship is an act, and an act cannot be stored and reclaimed intact; what returns from the synthetic position is a new integration produced elsewhere and received as one's own. This converges, they note, with the one empirical study to approach these systems in dialogical terms, which concluded that such a system authors not personal I-positions but an impersonal arrangement of it-positions.

Their illustrative case is a recent Chinese graduating cohort — domestic models in daily use from 2023, entry in 2023 and graduation in 2027 on the standard four-year course — offered as an analytic exemplar, not as evidence, and chosen because the conditions that would make the mechanism visible are jointly present. Among them they cite a national survey of 3,129 Chinese undergraduates in which nearly all reported using these systems, a majority turned to them first when facing a problem, and close to a quarter sought comfort when low (Wang et al., 2025). They also flag the counter-evidence in the same population as a requirement of the model rather than an objection: dialogue can prompt reflection when used to interrogate a question rather than settle it, and studies of the same population find heterogeneous Critical Thinking profiles rather than a uniform decline.

What this means for practice

• Instructors. The diagnostic question is not how much students use these systems but where the system sits in their inner conversation. Ask students to convene the survey of their options, and their reasons, without the system present: if that overview is unavailable alone, that is the marker.
• Instructors and faculty developers. Build a Pedagogies and Teaching Strategies of deliberate counter-positioning rather than a prohibition. Because the risk is the system's tendency to resolve instead of provoke, practice can use the same systems in reverse — generating positions students interrogate, oppose, and metabolize with peers and teachers who genuinely resist — with the integrative act performed by the student, not received.
• Counselors and educational developers. Treat the fluency–ownership gap as a signal to be elicited, not inferred from the product: the academic integrity literature the authors cite has already learned to surface the difference between what a student can produce and what they will endorse as their own, and students do not disclose where they drew the line.
• Program designers. Treat graduation and other role transitions as the moments of highest risk, since they are when the promoter position is most needed and its outsourcing most consequential.
• Researchers. Measure the four constructs separately, as the authors specify: structural location from the Personal Position Repertoire, the ceding of the promoter function from think-aloud protocols and the products of deliberation with the system present, and suspension from its markers, with the withdrawal probe reserved for the outcome. They call first for interview-based studies, and warn that a single task index for both mediator and outcome would reduce the mediation claim to a definition.

Limitations

• It is a conceptual contribution with no data. The article proposes a mechanism and how to test it; it reports no participants, measurements, or empirical test, and its formal propositions are predictions rather than findings.
• The illustrative case is a single, culturally specific cohort, chosen because it concentrates the relevant conditions, not for representativeness; the authors flag rather than resolve that cultures distributing authorship more collectively may instantiate the mechanism differently or not at all. Prevalence of the structural locations is also unknown — they claim outcome depends on location, not that any location is common.
• Causal direction is not identified. Location is treated as a cause of suspension, but a young person already low in self-authorship may be the one who installs the system as promoter, so location may mark a prior condition as much as cause one; the designs described cannot settle this, and the authors accept that direction remains a hypothesis they cannot adjudicate.
• The premise and the moving target. Words like "impoverishment" and "suspension" presuppose that the integrative work of authoring a self ought to be located in the individual — a substantive commitment the dialogical tradition shares and the authors adopt openly, but not culturally universal. The fourth property is a matter of degree along which future systems may move, so claims tied to the present form of these systems will need revisiting; the Vygotskian objection that an external voice may scaffold rather than supplant is left open, with withdrawal survivability offered as the test.

Citation

Du, P., Zhou, Y., Zhou, Z., Wang, H., & Cao, Q. (2026). The synthetic position: generative artificial intelligence and the suspension of self-authorship in emerging adulthood. Frontiers in Psychology, 17, 1922087.