Source-linked AI summary

The AI Adaptation Gap in Higher Education: Students, Faculty, and Administrative Staff

Yuriy S. Braun, Salavat M. Khafizov

arXiv:2608.25063v1cs.CY

TL;DR

Higher-education research has limited empirical comparison of how students, faculty, and administrative staff use and view AI. This study compares these groups at a large teacher-education university and finds a pronounced adaptation gap, with students reporting greater use and usefulness while staff report stronger integrity concerns and responsible-use norms.

  • Problem

    Empirical research rarely directly compares students, faculty, and administrative staff, despite their distinct roles in shaping university AI practice.

  • Method

    A cross-sectional survey compares AI use and attitudes across three university groups, models associations among key indices, and explores student heterogeneity through clustering.

  • Results

    Students reported higher AI-use intensity and perceived usefulness, while faculty and administrative staff reported stronger integrity concerns and responsible-use norms; usefulness had the strongest association with trust (beta = 0.402).

  • Takeaways & Limitations

    The findings support stakeholder-aware, task-sensitive AI policies that address academic integrity and transparent disclosure rather than applying blanket permitted-versus-prohibited rules.

  • Takeaways & Limitations

    Because the study is cross-sectional, its observed associations do not establish causal direction.

Abstract

from arXiv · show

The purpose of this study was to analyze patterns of artificial intelligence (AI) use and attitudes toward AI among students, faculty, and administrative staff at a large university specializing in teacher education. The analytical sample comprised 1809 students, 250 faculty members, and 62 administrative staff members (N = 2121). Three role-adapted 75-item questionnaires covered the frequency and contexts of AI use, perceived usefulness, trust and control, academic integrity concerns, responsible-use norms, institutional policy clarity, and perceived improvement in output quality. Data analysis included descriptive statistics, Welch group comparisons, pooled ordinary least squares (OLS) models, reliability and dimensionality checks for observed indices, and exploratory student-only K-means clustering. The results revealed a pronounced AI adaptation gap across university groups. Students reported higher current AI-use intensity and perceived usefulness than faculty and administrative staff, whereas faculty and administrative staff reported stronger academic integrity concerns and greater endorsement of responsible-use norms. In the pooled OLS trust model, perceived usefulness had the strongest standardized positive association with trust in AI (beta = 0.402); institutional policy clarity also had a positive but weaker association (beta = 0.223). Students reported higher perceived policy clarity than faculty, while neither group differed significantly from administrative staff. Exploratory clustering indicated heterogeneity among students in experience, competence, usefulness, trust, and control, but did not establish a latent typology across university groups. The cross-sectional, self-reported data show associations and group differences rather than causal effects on learning or objective outcomes.

1. Introduction

Research on AI in higher education has developed faster in theory than in empirical analysis of real-world use across students, faculty, and administrative staff. This study addresses that gap by examining AI-use contexts, associated perceptions and norms, group differences, regression associations, and exploratory student heterogeneity at a large teacher-education university.

  • Research gap: Research remains uneven, with empirical studies focusing mainly on students or faculty rather than real-world AI-use contexts across all three university groups.Broader stakeholder studies are emerging, but the literature remains less developed than theoretical discussion.
  • Study contribution: The study treats AI as embedded across assignments, communication, planning, assessment, administration, and service practices rather than as a single tool.It examines frequency of use alongside experience, competence, perceived usefulness, academic integrity concerns, trust, and institutional rules.
  • Analytical framework: AI-use and trust patterns are examined through perceived usefulness, self-reported competence, academic integrity concerns, and institutional policy clarity.The framework links user initiative with institutional rules, signals, and boundaries of permissible use.
  • Study objectives: The analysis compares reported AI-service use, weekly use, and current AI-use intensity among students, faculty, and administrative staff.It also compares perceived usefulness, academic integrity concerns, and institutional policy clarity across groups.
  • Study objectives: Regression models estimate associations among perceived usefulness, institutional policy clarity, trust, and current AI-use intensity, while K-means clustering explores student heterogeneity.The research questions additionally consider responsible-use norms and academic integrity concerns in pooled OLS models.

2. Theoretical Framework

The framework examines AI adoption through perceived usefulness, institutional risk, policy clarity, and responsible-use norms across students, faculty, and administrative staff. It treats these relationships as group-sensitive associations, while recognizing that self-reported, cross-sectional measures do not establish causal or objective educational effects.

  • Perceived usefulness: Perceived usefulness concerns whether AI makes educational or administrative activity more efficient, effective, or convenient, and is expected to relate positively to trust and use intensity.The framework emphasizes faster information processing, improved feedback quality, and reduced routine workload, while noting that causal direction cannot be established.
  • AI-use tensions: Institutional risk is distinguished from information inaccuracy: the analysis operationalized inappropriate use in assessment, independent work, and academic integrity, not privacy or reliability risks.This distinction separates institutionally inappropriate conduct from inaccurate or unreliable information risks excluded from the main models.
  • Policy clarity: Clear AI policies can define permitted uses, disclosure requirements, and assessment implications, and their association with trust is examined without assuming a directional causal effect.Policy clarity is framed differently across roles: procedural for administrative staff, pedagogical and ethical for faculty, and conduct-oriented for students.
  • Responsible use: Responsible use distinguishes the quality of AI use from its frequency and includes disclosure, explanation of AI-assisted steps, accuracy checking, and recognition of model limitations.Responsible-use norms, trust, and perceived quality improvement are self-reported and do not measure objective academic performance or establish educational effects.
  • Cross-group framework: The study expects use intensity, usefulness, trust, risks, and normative attitudes to differ across students, faculty, and administrative staff, requiring methods robust to unequal group sizes.Estimates for the administrative group require particular caution because the group sizes are markedly imbalanced.

3. Method

The study used a cross-sectional survey of 2121 respondents at a large teacher-education university, with role-adapted questionnaires measuring AI use, attitudes, norms, and institutional conditions. Analyses combined descriptive and Welch group comparisons, pooled OLS models, reliability and dimensionality checks, and exploratory student-only K-means clustering.

  • Study design and sample: 2121 respondents comprised 1809 students, 250 faculty members, and 62 administrative staff members in the final analytical datasets.The study used three cleaned, role-specific datasets from a large university specializing in teacher education.
  • Measures: Each role-adapted questionnaire contained 75 items covering AI-use experience, frequency, usefulness, trust, control, policy clarity, academic integrity, and responsible-use norms.Questionnaire wording was adapted to learning, teaching, or administrative activities, while block structures remained comparable.
  • Measures: Current AI-use intensity, perceived usefulness, trust, control, policy clarity, responsible AI-use norms, and academic integrity concerns were calculated from specified role-specific item sets.Indices used respondent-level means of available valid items; the academic integrity concerns index was an exact subset of the responsible AI-use norms index.
  • Statistical analysis: Welch’s one-way ANOVA and Bonferroni-adjusted Welch t-tests assessed group differences, with classical eta squared and Hedges’ g as effect sizes.Initial analyses also calculated frequencies, proportions, means, standard deviations, and 95% confidence intervals.
  • Statistical analysis: Two pooled OLS models estimated associations with trust and current AI-use intensity on a complete-data subsample of N = 1254, using administrative staff as the reference category.Continuous dependent variables and predictors were standardized, and predictor-by-group interactions were not estimated.
  • Exploratory clustering: Exploratory K-means clustering was conducted only among students using a complete-case subsample of n = 961 and a standardized four-cluster solution.The primary specification used nine features, fixed random seed 42, and ten initializations.

4. Results

Results showed a pronounced AI adaptation gap: students reported greater use intensity and usefulness, while faculty and administrative staff reported stronger integrity concerns and responsible-use norms. Associations, reliability diagnostics, experience patterns, and student-only clustering further revealed important measurement and interpretation limits.

  • AI use: 44.05% of students reported weekly AI use, compared with 21.60% of faculty and 15.79% of administrative staff, despite administrative staff having the highest overall-use prevalence at 85.48%.Overall AI-service use was reported by 1748 of 2121 respondents (82.41%); prevalence was 83.08% among students and 76.80% among faculty.
  • Group comparisons: Students had higher current AI-use intensity and perceived usefulness, whereas faculty and administrative staff had stronger academic integrity concerns and responsible-use norms.For current AI-use intensity, student differences versus faculty and administrative staff were g = 0.79 and g = 0.98, respectively; usefulness differences were g = 0.75 and g = 0.54.
  • Pooled OLS models: In the pooled trust model, perceived usefulness had the strongest association with trust (𝛽= 0.402), while policy clarity was positive but weaker (𝛽= 0.223).The model explained 40.8% of trust variance (N = 1254, R² = 0.408, adjusted R² = 0.404).
  • Experience and usefulness: Among students, perceived usefulness increased from 2.78 among non-users to 4.06 among those with more than two years’ experience, with Spearman’s r = 0.327, p < 0.001, n = 1454.The pattern was associative rather than causal, and self-selection remained possible.
  • Student clustering: Four student-only K-means clusters in n = 961 showed heterogeneity in experience, competence, usefulness, trust, and control, but did not establish stable psychological types across university groups.Cluster sizes were 221, 111, 378, and 251; the solution excluded faculty and administrative staff and was sensitive to features and parameters.

5. Discussion

The discussion interprets AI as an established university practice requiring policy clarity, task-sensitive governance, differentiated support, and attention to administrative processes. It also frames usefulness, integrity concerns, clarity, and trust as interconnected observed associations rather than causal effects.

  • Institutional policy implications: Most respondents reported using at least one AI service, and students reported weekly use more often than faculty and administrative staff while perceiving AI as more useful.The discussion argues that policy should address practices already developed rather than rely primarily on assumptions about future adoption.
  • Institutional policy implications: Perceived policy clarity was positively associated with trust in AI, although perceived usefulness had the larger standardized coefficient in the estimated model.The association should be interpreted alongside user experience, competence, usefulness, and academic integrity concerns; the cross-sectional data do not establish causality.
  • Institutional policy implications: Faculty and administrative staff reported AI use alongside stronger academic integrity concerns and greater endorsement of responsible-use norms than students.The discussion therefore calls for task-sensitive rules rather than a blanket distinction between permitted and prohibited uses.
  • Differentiated support: Student clustering indicated heterogeneity in experience, competence, usefulness, trust, and control, suggesting that one training and regulation approach may not serve all students equally.The exploratory clusters should not be treated as fixed psychological categories or used to classify individual students.
  • Administrative processes: Administrative staff are stakeholders in the university AI environment because they connect policy with everyday procedures, service regulations, internal guidance, and workflows.The data do not show that administrative practices cause AI adoption, but they are consistent with considering administrative processes alongside teaching and learning.
  • Conceptual interpretation: Perceived usefulness, academic integrity concerns, policy clarity, and trust form an interconnected context for AI use, but these propositions remain interpretations of observed associations.The discussion presents them as a basis for further research rather than evidence of causal pathways.

6. Limitations

The study’s single-university, cross-sectional, self-reported design limits representativeness, causal inference, measurement equivalence, and interpretation of group differences. Unequal samples, analysis-specific missing-data rules, exploratory clustering, and absent behavioral or objective outcomes further constrain conclusions, while the design remains useful for internal monitoring and future research.

  • Sampling and design: The single-university sample lacks cross-institutional representativeness, and undocumented recruitment and response information prevents estimating selection or nonresponse bias.Findings should not be generalized beyond the study setting without further evidence.
  • Sampling and design: Because the study was cross-sectional, observed associations do not establish causal direction and may reflect experience effects or user self-selection.Longitudinal designs are needed to distinguish these possibilities.
  • Measurement: Self-reports may contain social desirability bias, attention checks were absent, and limited-item diagnostics indicated that common method variance warrants consideration.These concerns are especially relevant to academic integrity and inappropriate AI-use measures.
  • Analysis constraints: 62 administrative respondents and reliability samples of 28–54, falling to 25 for the nine-item PCA diagnostic, limit precision for staff-group conclusions.The substantially unequal group sizes make administrative-staff estimates particularly uncertain.
  • Analysis constraints: Different missing-data rules produced varying analysis samples, while the primary K-means specification required complete responses on all original features.The OLS subsample was complete on calculated index scores, not necessarily every contributing item.
  • Measurement and clustering: The Tucker value of 0.934 for students and administrative staff was not a formal invariance test, so role-adapted indices should not be treated as fully equivalent latent constructs.The four-cluster K-means solution was student-only, used separate coding, lacked demonstrated optimality or stability, and was not a latent classification.
  • Model interpretation: Overlapping responsible-use and integrity indices, unadjusted conventional OLS errors, and unestimated group interactions preclude substantive interpretation of individual coefficients or slope differences.The overlapping indices share items 6A.3 and 6A.4.
  • Outcome measurement: Without behavioral traces, assignment artifacts, or objective performance measures, perceived quality improvement is subjective; the questionnaire also lacked a cognitive-delegation scale.The study therefore does not measure educational effects directly.

7. Conclusion

AI use was widespread but uneven across the university: students reported greater use intensity and usefulness, while faculty and administrative staff expressed stronger integrity concerns and responsible-use norms. Policy clarity was positively associated with trust, but the conclusions remain limited to associations and group differences within one university sample.

  • AI adaptation gap: Students reported higher current AI-use intensity and perceived usefulness, while faculty and administrative staff reported stronger academic integrity concerns and responsible-use norms.
  • Trust and policy: Perceived usefulness had a stronger standardized association with trust in AI than institutional policy clarity in the pooled OLS model.
  • Trust and policy: Students perceived greater policy clarity than faculty, while neither group differed significantly from administrative staff.
  • Limitations: The conclusions indicate statistical associations and group differences within one university sample, not causal effects or a stable latent typology of university stakeholders.

Appendix A. Structure of the Role-Adapted 75-Item Questionnaires

Each respondent group completed a role-adapted 75-item questionnaire with a comparable block structure covering demographics, AI use, usefulness, trust and control, policy, integrity, and social norms.

  • Questionnaire structure: 75 items comprised each role-adapted questionnaire administered to students, faculty, and administrative staff.The questionnaires used comparable block structures across respondent groups.
  • Questionnaire structure: The blocks assessed demographics and AI-use experience, policy awareness and access, current and expected AI-use frequency, and perceived usefulness.These domains occupied Blocks 0–3, totaling 23 items.

Declarations · Ethics approval and consent to participate

The rectorate approved the study protocol and ethics procedures under the university’s procedures for research involving human participants. All participants were older than 16 years and provided informed consent before completing the questionnaire; the institution remains anonymized.

  • Ethics approval and consent to participate: The rectorate served as the institutional approval body for research involving human participants.
  • Ethics approval and consent to participate: The rectorate approved the study protocol.
  • Ethics approval and consent to participate: The rectorate approved the study’s ethics procedures.
  • Ethics approval and consent to participate: All participants were older than 16 years.
  • Ethics approval and consent to participate: Informed consent was obtained before questionnaire completion.
  • Declarations: The institution remains anonymized in the manuscript.

Consent for publication … Use of generative AI and AI-assisted technologies

The study received no specific external grant, and the authors reported their respective contributions. Generative and AI-assisted tools supported several manuscript-related tasks, while the authors reviewed and edited the content and retained responsibility.

  • Funding: No specific grant supported the research from public, commercial, or not-for-profit sectors.
  • Authors’ contributions: Yuriy S. Braun contributed to conceptualization, methodology, formal analysis, investigation, drafting, revision, editing, and supervision.
  • Authors’ contributions: Salavat M. Khafizov contributed to investigation, data curation, and validation.
  • Authors’ contributions: Salavat M. Khafizov also contributed to resources, project administration, and manuscript review and editing.
  • Authors’ contributions: Both authors read and approved the final manuscript.
  • Use of generative AI and AI-assisted technologies: Generative and AI-assisted tools supported translation, draft wording, organization, and language and terminology review.
  • Use of generative AI and AI-assisted technologies: The authors reviewed and edited the content and take full responsibility for the manuscript.
Loading 2608.25063v1…