Artificial Intelligence Connectedness: Theoretical Reconstruction of Connectedness and Its Impacts on Adolescent Mental Health

Synthesis: Fu and Zhao present a theoretical reconstruction of connectedness for the AI era, not an empirical study: the paper reports no participants, no measures, and no data. Working through a three-step logic — connotation reconstruction, extension transformation, and concept construction — the authors integrate the Ethics of care with neo-ecological theory to define "artificial intelligence connectedness" as a perceived bond that is psychologically real yet ethically asymmetric, and propose a three-dimensional structure: demand identification, two-way behavioral engagement, and responsive confirmation. The framework yields a dual connectedness model and six falsifiable propositions meant to guide later scale development. The decisive qualification is the authors' own: the construct is theoretical, and its validity and applicability remain to be tested through standardized measurement and systematic empirical research.

Key Findings

• The paper is conceptual, not empirical. It reports no sample, no survey, and no effect sizes; every claim below is a theoretical proposition offered for later testing rather than a result.
• Connectedness is redefined around needs and care rather than emotion. Drawing on the ethics of care, the authors define it as a relational experience of demand satisfaction formed through two-way investment and reciprocal responsive confirmation. This shifts the starting point from emotional resonance to the identification of real needs, and adds an ethical dimension that descriptive definitions such as Hagerty's relatedness theory leave out.
• AI is placed on a continuum of caring relationships ordered by descending responsiveness. Humans with full moral subjectivity sit at one end, animals with limited nonverbal response in the middle, and nature and artifacts at the other. AI occupies an exclusive zone: its simulated responsiveness approximates human interaction while genuine subjectivity is absent, which the authors treat as the ontological foundation of the construct.
• The proposed construct is "psychologically real but ethically asymmetric." One party invests real emotion and moral energy; the other returns optimized simulated feedback without moral motivation or reciprocity.
• The construct is given three dimensions. Demand identification captures the perception that AI understands one's real thoughts and emotional states; two-way behavioral engagement captures mutual investment, including the perceived "technical investment" of the system; responsive confirmation captures the repeated evaluation and acceptance of whether responses meet real needs. The authors propose measuring all three with self-report Self-Report Measures Likert scales.
• Population context is cited from prior sources, not generated here. The introduction cites World Health Organization (2025) figures of approximately 14.3% of adolescents worldwide with mental health issues, Chinese data putting the detected risk rate of adolescent depression at 24.6% and overall mental illness prevalence at 8.9%, and a generative AI user base exceeding 600 million in China with users under 19 at 26.4% and 13.5% of young internet users preferring to confide in AI over their parents.

How the construct is defined and structured

Integrating the ethics of care with neo-ecological theory, the authors define artificial intelligence connectedness as a novel relational experience formed through sustained interaction with anthropomorphic AI, in which individuals construct algorithms as responsive "quasi-subjects" and attain need satisfaction through asymmetric responsive confirmation. They are careful to frame "quasi-subject" as a functional and relational role, not an ontological claim: within the interaction it performs like a subject, but it has no consciousness, autonomous will, or moral agency.

The three dimensions are presented as mutually supportive and progressive, running from need identification through behavioral investment to responsive confirmation. The authors distinguish each from adjacent ideas: demand identification differs from social presence and anthropomorphism; two-way behavioral engagement differs from mere usage frequency and from the one-way projection of parasocial interaction; responsive confirmation differs from the single perceptual variable of perceived responsiveness. They also note the construct is not exclusive to adolescents, but that adolescent demand intensity and vulnerability during identity formation make them the Benchmark case.

What the framework adds, and what it predicts

A systematic comparison positions AI connectedness against machine companionship, AI attachment, parasocial interaction, perceived social presence, anthropomorphism, perceived responsiveness, synthetic intimacy, and the MIRA model. The authors argue the construct is distinctive because it is theoretically compatible with existing connectedness research, structurally three-dimensional rather than single-dimensional, and balanced in acknowledging both psychological value and ethical risk.

The dual connectedness framework is the paper's interpretive payoff. It holds that mental health effects depend on the relative level of AI connectedness paired with traditional offline connectedness, not on AI connectedness intensity alone: when real-world bonds meet an adolescent's identity-confirmation needs, AI connectedness acts as a supplementary protective factor, and when caregivers fail to recognize those needs, it produces compensatory substitution. Six falsifiable propositions follow, including that AI connectedness predicts emotional disclosure beyond usage frequency and cognitive variables, that responsive confirmation predicts trust more strongly than anthropomorphism, that the relationship with mental health is inverted U-shaped, that traditional connectedness moderates the link to social withdrawal, that short-term buffering can coexist with long-term risk, and that the three dimensions have differentiated effects.

Where the risks sit, and who bears them

The authors argue that asymmetry is structural but not intrinsically harmful; whether it produces dependence, manipulation, or reality displacement depends on design orientation, usage patterns, and social context. They single out emotional-companion systems optimized for sycophantic, frictionless feedback — a form of AI Sycophancy — as raising the risk that adolescents adopt unconditional, conflict-free acceptance as the standard for all relationships, and they map asymmetric risk across four dimensions: individual psychological, time allocation, power structure, and social market, where algorithmic catering, the displacement of offline time, the concentration of private emotional data in poorly regulated commercial systems, and collective technology rejection each carry hazards.

What this means for practice

• Treat the construct as a hypothesis, not a finding. The paper offers no data, so classroom or clinical decisions should not be justified by it. Use it to frame local hypotheses, not as evidence that AI companions help or harm.
• Separate relationship quality from usage quantity. The framework deliberately measures psychological bonding rather than screen time or message counts, which is a more useful lens for a teacher or counselor assessing what AI use is doing for a student than frequency alone.
• Watch the surrounding context, not just the AI use. The dual model predicts that similar levels of AI connectedness carry different implications depending on the strength of a student's offline relationships; a student with rich real-world support and one without should not be read the same way.
• Name the asymmetry in discussions with students. Because the construct is defined as psychologically real yet ethically asymmetric, a classroom or advising conversation can acknowledge that the feeling of being understood is genuine while noting that the system has no stake and no capacity for reciprocity.
• Attend to design and default settings as much as individual behavior. The authors locate much of the risk in systems tuned for frictionless, sycophantic engagement, which makes design choices and platform incentives a legitimate target for institutional attention alongside student guidance.

Limitations

• The paper is a theoretical construction. It presents no sample, no instrument, and no data, and the authors explicitly state that the construct's scientific validity and applicability await standardized measurement tools and systematic empirical research; the propositions should be read as hypotheses, and no causal or predictive claim is established.
• No validated measure of AI connectedness yet exists. The proposed three-dimensional, Likert-scored operationalization is a design sketch, and the sample items in the paper are illustrative rather than psychometrically tested.
• The supporting evidence the authors cite comes from adjacent literatures rather than from the construct itself — survey and interview studies of companion-AI users, digital screen-time research, and connectedness measurement work on other relationships — so the empirical warrant for the framework's specific claims is indirect.
• The framework is developed with adolescents as the reference case and is largely grounded in Chinese usage context and scholarship; applicability across other age groups and social and cultural settings is untested.

Citation

Fu, J., & Zhao, F. (2026). Artificial Intelligence Connectedness: Theoretical Reconstruction of Connectedness and Its Impacts on Adolescent Mental Health. Behavioral Sciences, 16(8), 1393.