Psychological Mechanisms of Generative AI Discontinuance Intention among Chinese K-12 Teachers

Huimin He; Qian Chen; Yiran Du

arxiv: 2605.16648 · v1 · pith:FPL3RUHTnew · submitted 2026-05-15 · 💻 cs.HC

Psychological Mechanisms of Generative AI Discontinuance Intention among Chinese K-12 Teachers

Yiran Du , Qian Chen , Huimin He This is my paper

Pith reviewed 2026-05-20 15:36 UTC · model grok-4.3

classification 💻 cs.HC

keywords generative AIdiscontinuance intentionK-12 teachersAI anxietyuser satisfactionprivacy concernalgorithmic opacitystructural equation modeling

0 comments

The pith

Privacy concerns, algorithmic opacity, and hallucinations raise AI anxiety that increases Chinese K-12 teachers' plans to stop using generative AI, while perceived intelligence, personalization, and interactivity raise satisfaction that low

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests how Chinese K-12 teachers evaluate generative AI and what that does to their feelings and plans. Cognitive judgments about risks like privacy leaks, hidden algorithms, and false outputs create anxiety that pushes teachers toward quitting the tools. Positive judgments about the AI seeming smart, tailored, and responsive create satisfaction that pulls teachers toward continued use. The authors apply a thinking-feeling-acting sequence to survey data from 256 teachers and map several combinations of factors that together produce strong quitting intentions. Readers who work with AI in schools would care because the results point to concrete levers that could keep teachers engaged rather than driving them away.

Core claim

Using the Cognition-Affect-Conation framework, the study demonstrates that privacy concern, algorithmic opacity, and information hallucination elevate AI anxiety, which strengthens discontinuance intention, whereas perceived intelligence, perceived personalisation, and perceived interactivity elevate satisfaction, which weakens discontinuance intention. Fuzzy-set qualitative comparative analysis identifies multiple distinct configurations of these risks, affordances, anxiety, and satisfaction that each produce high discontinuance intention.

What carries the argument

The Cognition-Affect-Conation framework, which sequences cognitive evaluations of generative AI into affective states of anxiety or satisfaction and then into behavioral discontinuance intention.

Load-bearing premise

The survey answers given by the 256 teachers accurately capture their stable views on generative AI and that the measured links run from the listed perceptions through anxiety or satisfaction to quitting plans rather than the reverse or from unmeasured causes.

What would settle it

A longitudinal panel study that measures the same teachers' anxiety, satisfaction, and discontinuance intention at multiple time points and finds that changes in anxiety or satisfaction reliably precede rather than follow changes in discontinuance intention.

Figures

Figures reproduced from arXiv: 2605.16648 by Huimin He, Qian Chen, Yiran Du.

read the original abstract

This study examines the psychological mechanisms underlying Chinese K-12 teachers' discontinuance intention toward generative AI. Drawing on the Cognition-Affect-Conation framework, the study investigates how cognitive evaluations of generative AI shape affective responses and subsequently influence behavioural intention. Survey data from 256 Chinese K-12 teachers were analysed using structural equation modelling and fuzzy-set qualitative comparative analysis. The results showed that privacy concern, algorithmic opacity, and information hallucination increased AI anxiety, which in turn strengthened discontinuance intention. Conversely, perceived intelligence, perceived personalisation, and perceived interactivity enhanced satisfaction, which reduced discontinuance intention. The configurational analysis further identified multiple pathways leading to high discontinuance intention, highlighting the combined roles of technological risks, AI anxiety, weak affordance perceptions, and low satisfaction. These findings extend research on post-adoption generative AI use in education and suggest that sustainable integration requires both reducing technological uncertainty and enhancing teachers' positive user experiences.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Standard cross-sectional SEM plus fsQCA on 256 Chinese K-12 teacher surveys maps GenAI risks to anxiety and affordances to satisfaction, but directionality stays unproven.

read the letter

This paper applies the Cognition-Affect-Conation model to generative AI discontinuance among Chinese K-12 teachers. It reports that privacy concerns, opacity, and hallucination raise anxiety and thus discontinuance intention, while perceived intelligence, personalization, and interactivity raise satisfaction and lower it. The fsQCA part identifies several configurations that combine these elements into high discontinuance paths. That dual-method setup on a relevant population is the main concrete addition here, and the choice of a non-Western teacher sample gives the results some local grounding that prior Western-heavy work lacks. The practical pointers on reducing uncertainty and boosting positive experiences follow directly from the coefficients and configurations they present. The central limitation is the one the stress-test flags. A single-wave self-report survey cannot separate the proposed causal directions from reverse effects or common-method bias. Teachers already leaning toward quitting could simply rate the technology more negatively on the risk items. The paper does not appear to include longitudinal measures, instruments, or lagged designs that would tighten the mechanism claims. Sample details and validation steps for the scales are not visible in the abstract, so those sections will need close checking for representativeness and bias controls. This work sits squarely in the education-technology adoption literature. Readers who study teacher use of AI tools or who run professional development programs could extract usable factors from the risk and affordance lists. It is coherent on its own terms and shows normal engagement with the framework and methods, so it clears the bar for peer review. A referee would likely press on the causality point and ask for more on measurement quality, but the core empirical exercise is worth the time.

Referee Report

2 major / 2 minor

Summary. The manuscript claims that cognitive evaluations of generative AI shape affective responses and thereby influence discontinuance intention among Chinese K-12 teachers, per the Cognition-Affect-Conation framework. Survey data from 256 teachers analyzed via SEM show that privacy concern, algorithmic opacity, and information hallucination raise AI anxiety and thus discontinuance intention, while perceived intelligence, personalisation, and interactivity raise satisfaction and thus lower it. fsQCA identifies multiple configurations combining technological risks, anxiety, weak affordances, and low satisfaction as pathways to high discontinuance intention.

Significance. If the directional mechanisms hold, the work extends post-adoption research on generative AI in education by integrating risk and positive-affordance perspectives and offers practical guidance for sustainable integration. The dual-method approach (SEM for net effects plus fsQCA for configurations) is a clear strength that supplies complementary insights from the same dataset.

major comments (2)

[Results (SEM paths) and Discussion] Results (SEM paths) and Discussion: The central claim interprets the paths (privacy concern/algorithmic opacity/information hallucination → AI anxiety → discontinuance intention; and the parallel satisfaction path) as evidence of directional psychological mechanisms. The single-wave cross-sectional survey cannot distinguish these from reverse causation or common-method bias, since teachers already inclined to discontinue may retroactively rate the technology more negatively.
[Methods and fsQCA section] Methods and fsQCA section: The configurational solutions for high discontinuance intention rest on the identical cross-sectional self-report measures; they therefore do not independently establish the temporal ordering or causal precedence assumed in the Cognition-Affect-Conation model.

minor comments (2)

[Abstract and Methods] The abstract and methods description omit explicit reporting of measurement validation (reliability, convergent/discriminant validity), common-method bias diagnostics, and sample representativeness, all of which are needed to evaluate the soundness of the SEM and fsQCA results.
[Methods] Notation for the constructs (e.g., exact item wording or scale anchors for 'algorithmic opacity' and 'information hallucination') should be provided in a table or appendix to allow replication.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments. These observations correctly identify a core methodological constraint of our cross-sectional design. We address each point below and have revised the manuscript to more explicitly acknowledge the limits on causal inference while preserving the theoretical grounding of the Cognition-Affect-Conation framework.

read point-by-point responses

Referee: Results (SEM paths) and Discussion: The central claim interprets the paths (privacy concern/algorithmic opacity/information hallucination → AI anxiety → discontinuance intention; and the parallel satisfaction path) as evidence of directional psychological mechanisms. The single-wave cross-sectional survey cannot distinguish these from reverse causation or common-method bias, since teachers already inclined to discontinue may retroactively rate the technology more negatively.

Authors: We agree that the cross-sectional, single-wave survey cannot establish temporal precedence or rule out reverse causation and common-method bias. The directional hypotheses are derived from the Cognition-Affect-Conation framework rather than from the data alone. In the revised manuscript we have added a dedicated paragraph in the Limitations section that states this constraint plainly, notes that the observed associations are consistent with the theoretical model, and recommends longitudinal or experimental designs for future work. No causal language has been strengthened; the discussion now consistently uses terms such as “associated with” and “consistent with the framework.” revision: yes
Referee: Methods and fsQCA section: The configurational solutions for high discontinuance intention rest on the identical cross-sectional self-report measures; they therefore do not independently establish the temporal ordering or causal precedence assumed in the Cognition-Affect-Conation model.

Authors: We concur that the fsQCA solutions, like the SEM results, rest on the same cross-sectional self-report data and therefore cannot independently demonstrate temporal ordering. The fsQCA analysis is presented as a complementary, exploratory technique that identifies equifinal configurations rather than as a causal test. The revised manuscript now includes an explicit statement in both the Methods and Discussion sections clarifying that directional interpretations rest on the theoretical framework and that both analytical approaches share the limitations of the data collection design. Future research directions have been updated to emphasize the value of time-lagged or multi-wave studies. revision: yes

Circularity Check

0 steps flagged

Empirical survey study with no circular derivation steps

full rationale

The paper collects primary survey data from 256 Chinese K-12 teachers and applies standard structural equation modelling plus fsQCA to test directional paths in the pre-existing Cognition-Affect-Conation framework. All central claims (privacy/opacity/hallucination increasing anxiety and discontinuance; intelligence/personalisation/interactivity increasing satisfaction and reducing discontinuance) are statistical outputs from new observations rather than quantities defined by the paper's own equations or by self-citation chains. No fitted parameters are relabelled as predictions, no uniqueness theorems are imported from the authors' prior work, and no ansatz or renaming occurs. The derivation chain therefore remains independent of its own inputs and is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper relies on the Cognition-Affect-Conation framework as a domain assumption and standard survey-analysis techniques without introducing new free parameters or invented entities in the abstract.

axioms (1)

domain assumption The Cognition-Affect-Conation framework accurately structures how cognitive evaluations of generative AI shape affective responses and behavioral intentions
Invoked to organize the investigation of anxiety and satisfaction pathways.

pith-pipeline@v0.9.0 · 5690 in / 1307 out tokens · 55307 ms · 2026-05-20T15:36:22.715947+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Survey data from 256 Chinese K-12 teachers were analysed using structural equation modelling and fuzzy-set qualitative comparative analysis... privacy concern, algorithmic opacity, and information hallucination increased AI anxiety, which in turn strengthened discontinuance intention.
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The results showed that privacy concern, algorithmic opacity, and information hallucination increased AI anxiety...

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

[1]

Introduction Generative AI, has rapidly entered educational practice, offering new opportunities for teachers to generate instructional materials, design assessments, provide feedback, and support lesson preparation (Qian, 2025; Tan et al., 2025). In K–12 education, generative AI is increasingly viewed as a tool that may reduce teachers’ workload, improve...

work page 2025
[2]

Wang et al., 2025)

Literature Review 2.1 Generative AI in Teaching Generative AI refers to artificial intelligence systems capable of producing new content, such as text, images, or code, based on patterns learned from large datasets (Y . Wang et al., 2025). With the emergence of large language models and multimodal generative systems, generative AI has begun to influence e...

work page 2025
[3]

Theoretical Framework and Hypothesis Development 3.1 Cognition–Affect–Conation Framework The Cognition–Affect–Conation (CAC) framework explains behavioural intention as a sequential psychological process in which individuals’ cognitive evaluations influence emotional reactions, which subsequently shape behavioural responses (Zeng et al., 2023). The framew...

work page 2023
[4]

The Conceptual Model 3.2 AI Anxiety, Privacy Concern, Algorithmic Opacity, and Information Hallucination AI anxiety refers to users’ feelings of apprehension, uncertainty, or discomfort when interacting with artificial intelligence technologies (Johnson & Verdicchio, 2017). Prior research in human–computer interaction and automation has shown that individ...

work page 2017
[5]

Methods 4.1 Research Design This study adopted a quantitative, cross-sectional survey design to examine the psychological mechanisms underlying Chinese K–12 teachers’ discontinuance intention toward generative AI. Guided by the Cognition–Affect–Conation framework, the study tested a structural model in which teachers’ cognitive evaluations of generative A...

work page 2023
[6]

Du, 2024)

Variable Category Frequency (n) Percentage (%) Age Under 25 60 23.4 25–34 110 43.0 35–44 56 21.9 45 or above 30 11.7 Gender Male 110 43.0 Female 146 57.0 Teaching subject STEM 120 46.9 Non-STEM 136 53.1 Teaching experience Less than 3 years 70 27.3 3–5 years 64 25.0 6–10 years 68 26.6 More than 10 years 54 21.1 Teaching level Primary 90 35.2 Junior second...

work page 2025
[7]

Because the data were obtained entirely from a self-report questionnaire, common method variance (CMV) was examined (Podsakoff et al., 2024)

to assess item clarity and contextual appropriateness, and minor wording adjustments were made accordingly. Because the data were obtained entirely from a self-report questionnaire, common method variance (CMV) was examined (Podsakoff et al., 2024). Harman’s single-factor test indicated that the first unrotated factor explained 23.6% of the total variance...

work page 2024
[8]

Discussion 6.1 Net Effects of Enablers and Inhibitors on Generative AI Discontinuance Intention The SEM results provide clear support for the proposed Cognition–Affect–Conation logic in explaining Chinese K–12 teachers’ generative AI discontinuance intention. Specifically, AI anxiety exerted a positive effect on discontinuance intention, whereas satisfact...

work page 2023
[9]

watching-eye

Conclusion This study examined the psychological mechanisms underlying Chinese K–12 teachers’ generative AI discontinuance intention through the Cognition–Affect–Conation framework and a dual perspective of enablers and inhibitors. The SEM results showed that privacy concern, algorithmic opacity, and information hallucination increased discontinuance inte...

work page doi:10.1016/j.caeai.2023.100132 2023
[10]

https://doi.org/10.1057/s41599-024-03811-x Tan, X., Cheng, G., & Ling, M. H. (2025). Artificial intelligence in teaching and teacher professional development: A systematic review. Computers and Education: Artificial Intelligence, 8, 100355. https://doi.org/10.1016/j.caeai.2024.100355 Tang, M., Jia, K., He, H., Wang, C., Zou, B., & Du, Y . (2026). Acceptan...

work page doi:10.1057/s41599-024-03811-x 2025
[11]

https://doi.org/10.1016/j.ijer.2026.102963 Tsai, H

International Journal of Educational Research, 137, 102963. https://doi.org/10.1016/j.ijer.2026.102963 Tusseyeva, I., Sandygulova, A., & Rubagotti, M. (2024). Perceived intelligence in human-robot interaction: A review. IEEE Access, 12, 151348–151359. https://doi.org/10.1109/ACCESS.2024.3478751 Vaassen, B. (2022). AI, opacity, and personal autonomy. Philo...

work page doi:10.1016/j.ijer.2026.102963 2026
[12]

https://doi.org/10.1007/s13347-022-00577-5 Wang, C., Du, Y., & Zou, B. (2026). Learners’ acceptance and use of multimodal artificial intelligence (AI)‐generated content in AI‐mediated informal digital learning of English. International Journal of Applied Linguistics, 36(1), 927–940. https://doi.org/10.1111/ijal.12827 Wang, C., Zou, B., Du, Y., & Wang, Z. ...

work page doi:10.1007/s13347-022-00577-5 2026
[13]

https://doi.org/10.1186/s41239-023-00420-7 Zhang, W., Zou, B., & Du, Y . (2026). Teachers’ perceptions of the current practices and challenges in English for academic purposes: A survey study at universities in Shanghai, China. International Journal of English for Academic Purposes: Research and Practice, 6(1), 7–28. https://doi.org/10.3828/ijeap.2026.2 Z...

work page doi:10.1186/s41239-023-00420-7 2026

[1] [1]

Introduction Generative AI, has rapidly entered educational practice, offering new opportunities for teachers to generate instructional materials, design assessments, provide feedback, and support lesson preparation (Qian, 2025; Tan et al., 2025). In K–12 education, generative AI is increasingly viewed as a tool that may reduce teachers’ workload, improve...

work page 2025

[2] [2]

Wang et al., 2025)

Literature Review 2.1 Generative AI in Teaching Generative AI refers to artificial intelligence systems capable of producing new content, such as text, images, or code, based on patterns learned from large datasets (Y . Wang et al., 2025). With the emergence of large language models and multimodal generative systems, generative AI has begun to influence e...

work page 2025

[3] [3]

Theoretical Framework and Hypothesis Development 3.1 Cognition–Affect–Conation Framework The Cognition–Affect–Conation (CAC) framework explains behavioural intention as a sequential psychological process in which individuals’ cognitive evaluations influence emotional reactions, which subsequently shape behavioural responses (Zeng et al., 2023). The framew...

work page 2023

[4] [4]

The Conceptual Model 3.2 AI Anxiety, Privacy Concern, Algorithmic Opacity, and Information Hallucination AI anxiety refers to users’ feelings of apprehension, uncertainty, or discomfort when interacting with artificial intelligence technologies (Johnson & Verdicchio, 2017). Prior research in human–computer interaction and automation has shown that individ...

work page 2017

[5] [5]

Methods 4.1 Research Design This study adopted a quantitative, cross-sectional survey design to examine the psychological mechanisms underlying Chinese K–12 teachers’ discontinuance intention toward generative AI. Guided by the Cognition–Affect–Conation framework, the study tested a structural model in which teachers’ cognitive evaluations of generative A...

work page 2023

[6] [6]

Du, 2024)

Variable Category Frequency (n) Percentage (%) Age Under 25 60 23.4 25–34 110 43.0 35–44 56 21.9 45 or above 30 11.7 Gender Male 110 43.0 Female 146 57.0 Teaching subject STEM 120 46.9 Non-STEM 136 53.1 Teaching experience Less than 3 years 70 27.3 3–5 years 64 25.0 6–10 years 68 26.6 More than 10 years 54 21.1 Teaching level Primary 90 35.2 Junior second...

work page 2025

[7] [7]

Because the data were obtained entirely from a self-report questionnaire, common method variance (CMV) was examined (Podsakoff et al., 2024)

to assess item clarity and contextual appropriateness, and minor wording adjustments were made accordingly. Because the data were obtained entirely from a self-report questionnaire, common method variance (CMV) was examined (Podsakoff et al., 2024). Harman’s single-factor test indicated that the first unrotated factor explained 23.6% of the total variance...

work page 2024

[8] [8]

Discussion 6.1 Net Effects of Enablers and Inhibitors on Generative AI Discontinuance Intention The SEM results provide clear support for the proposed Cognition–Affect–Conation logic in explaining Chinese K–12 teachers’ generative AI discontinuance intention. Specifically, AI anxiety exerted a positive effect on discontinuance intention, whereas satisfact...

work page 2023

[9] [9]

watching-eye

Conclusion This study examined the psychological mechanisms underlying Chinese K–12 teachers’ generative AI discontinuance intention through the Cognition–Affect–Conation framework and a dual perspective of enablers and inhibitors. The SEM results showed that privacy concern, algorithmic opacity, and information hallucination increased discontinuance inte...

work page doi:10.1016/j.caeai.2023.100132 2023

[10] [10]

https://doi.org/10.1057/s41599-024-03811-x Tan, X., Cheng, G., & Ling, M. H. (2025). Artificial intelligence in teaching and teacher professional development: A systematic review. Computers and Education: Artificial Intelligence, 8, 100355. https://doi.org/10.1016/j.caeai.2024.100355 Tang, M., Jia, K., He, H., Wang, C., Zou, B., & Du, Y . (2026). Acceptan...

work page doi:10.1057/s41599-024-03811-x 2025

[11] [11]

https://doi.org/10.1016/j.ijer.2026.102963 Tsai, H

International Journal of Educational Research, 137, 102963. https://doi.org/10.1016/j.ijer.2026.102963 Tusseyeva, I., Sandygulova, A., & Rubagotti, M. (2024). Perceived intelligence in human-robot interaction: A review. IEEE Access, 12, 151348–151359. https://doi.org/10.1109/ACCESS.2024.3478751 Vaassen, B. (2022). AI, opacity, and personal autonomy. Philo...

work page doi:10.1016/j.ijer.2026.102963 2026

[12] [12]

https://doi.org/10.1007/s13347-022-00577-5 Wang, C., Du, Y., & Zou, B. (2026). Learners’ acceptance and use of multimodal artificial intelligence (AI)‐generated content in AI‐mediated informal digital learning of English. International Journal of Applied Linguistics, 36(1), 927–940. https://doi.org/10.1111/ijal.12827 Wang, C., Zou, B., Du, Y., & Wang, Z. ...

work page doi:10.1007/s13347-022-00577-5 2026

[13] [13]

https://doi.org/10.1186/s41239-023-00420-7 Zhang, W., Zou, B., & Du, Y . (2026). Teachers’ perceptions of the current practices and challenges in English for academic purposes: A survey study at universities in Shanghai, China. International Journal of English for Academic Purposes: Research and Practice, 6(1), 7–28. https://doi.org/10.3828/ijeap.2026.2 Z...

work page doi:10.1186/s41239-023-00420-7 2026