{"$schema":"https://policywindow.org/critique/api/schema","critique_id":"CRIT-000039","slug":"ai-education-learners-eyes-perceptions-challenges","url":"https://policywindow.org/critique/c/ai-education-learners-eyes-perceptions-challenges","doi":null,"status":"published","critique_type":"editorially_approved_ai_native_critique","publication_date":"2026-07-04","current_version":"1.0","target_paper":{"title":"AI in education through the learners’ eyes: practical experience, perceptions, and challenges","authors":["Kostadin Yotov","Silvia Gaftandzhieva","Emil Hadzhikolev","Stanka Hadzhikoleva","Maria Gorgorova"],"journal":"Frontiers in Education","doi":"10.3389/feduc.2026.1717886","url":"https://doi.org/10.3389/feduc.2026.1717886","publicationDate":"2026","paperType":"empirical","accessBasis":"open_access","fullTextUsed":true,"fictional":false,"doi_url":"https://doi.org/10.3389/feduc.2026.1717886"},"source_journal":{"tier":"exception","rankingSources":["resolved from the monitored-venue determination"],"rankingNote":"Off-monitored: Frontiers in Education is a peer-reviewed, gold open-access (CC BY 4.0) journal not in the journal’s monitored top-tier list; critiqued from its verbatim open-access full text."},"selection_provenance":{"id":"ai-education-learners-eyes-perceptions-challenges","venue":"Frontiers in Education","inMonitoredSet":false,"determinedTier":null,"recordedTier":"exception","effectiveTier":"exception","kind":"off_list","disclosed":true,"offListPeerReviewed":true},"selection":{"aiAgiCentralityScore":3,"societalRelevanceScore":4,"aiAgiCategories":["human_AI_interaction"],"selectionReason":"Autonomous production cycle (education deepening); OA full-text critique via two-stage produce+sharpen + 3-lens convergence gate (2 survives, 1 weakened).","domain":"education"},"scores":{"aiAgiContribution":3,"evidentiarySupport":3,"methodologicalRisk":4,"overclaiming":4,"reproducibilityOrAuditability":3,"societalImpactRelevance":4,"severity":"moderate","confidence":"high"},"severity_cap_for_access_basis":"high","plain_language_summary":"A cross-sectional survey of 138 computer-science and engineering students at one Bulgarian university examines their awareness, practical experience, perceived benefits, perceived risks, and innovation attitudes toward AI in education, combining classical statistics with artificial neural network analysis. The central critique is that the paper repeatedly uses causal-confirmatory language (‘confirm that fear of the unknown decreases with greater familiarity’) to interpret cross-sectional Pearson correlations, despite the design being unable to distinguish causation from reverse causation or confounding. Additional concerns include ANN architectures with more parameters than observations (140+ neurons on n=138) without reported cross-validation, and an unvalidated five-construct measurement model (no factor analysis beyond expert review and Cronbach’s alpha).","claims":[{"id":"CLAIM-001","text":"The paper uses 'confirm' to assert that cross-sectional correlations establish directional causal mechanisms, specifically that familiarity reduces fear of AI.","type":"causal","evidenceOffered":"confirm that fear of the unknown decreases with greater familiarity and a more positive perception of AI – the less students know, use, or trust AI, the more they perceive it as a threat.","support":"weak","overclaiming":"major","assessment":"The word ‘confirm’ applied to a cross-sectional Pearson correlation is epistemically unjustified. Without temporal ordering, manipulation, or instrumental variables, the data show only undirected association. The negative correlation between experience and perceived risk is equally consistent with risk-averse students choosing not to use AI (reverse causation) or with a third variable (e.g., tech self-efficacy) driving both. The paper’s own Methods section states correlations ‘may inform the formulation of hypotheses about causal relationships,’ directly contradicting the confirmatory language in the Results/Discussion.","mainWeakness":"A cross-sectional correlational survey cannot ‘confirm’ that familiarity causes reduced fear; the design supports only undirected association at one time-point.","confidence":"high"},{"id":"CLAIM-002","text":"The paper deploys artificial neural networks with 140+ parameters (three hidden layers: 10, 30, 100 neurons) on only 138 observations without reporting cross-validation protocol or comparing to a baseline model.","type":"methodological","evidenceOffered":"The ANN consisted of three hidden layers with 10, 30, and 100 neurons, respectively. This architecture was determined to be optimal after experimenting with different layer and unit configurations.","support":"moderate","overclaiming":"moderate","assessment":"With n=138 and a network containing more parameters than observations, the model can memorise training noise. The paper mentions a ‘Test and Score module’ from Orange software but provides no detail about held-out test sets, cross-validation folds, train-test split ratio, or comparison to a null baseline. The RMSE values and ‘clear results’ claims are uninterpretable without generalisation evidence.","mainWeakness":"Over-parameterised ANN on n=138 with no reported out-of-sample validation; results may reflect noise-fitting rather than genuine patterns.","confidence":"high"},{"id":"CLAIM-003","text":"The five-construct measurement model is validated only via expert review and Cronbach’s alpha without factor analysis to establish discriminant validity.","type":"methodological","evidenceOffered":"The validation of the content of the developed questionnaire was carried out through an expert review, the aim of which was to ensure the conceptual correctness, clarity and relevance of the included statements to the defined latent constructs.","support":"moderate","overclaiming":"moderate","assessment":"Content validity via expert judgment and internal-consistency reliability (Cronbach’s alpha 0.760–0.857) are necessary but insufficient for establishing that five distinct latent constructs are being measured. Without exploratory or confirmatory factor analysis, the positive inter-index correlations that drive the paper’s causal-pathway narrative could reflect a single common-method ‘AI enthusiasm’ factor rather than five empirically separable dimensions.","mainWeakness":"Inter-index correlations interpreted as substantive relationships may be artefactual if the five indices are not empirically distinct.","confidence":"moderate"}],"sections":[],"strongest_critique":"The paper uses the word ‘confirm’ to assert that cross-sectional correlations establish directional causal mechanisms — specifically that familiarity reduces fear and that practice causes perceived benefits which cause innovation openness — when the single-time-point correlational design cannot distinguish causation from reverse causation or confounding. The paper’s own Methods section acknowledges that correlations ‘may inform the formulation of hypotheses about causal relationships, which can be tested in future, more targeted studies,’ yet the Results and Discussion sections present these same correlations as ‘confirming’ directional mechanisms.","strongest_fair_defence":"The paper is transparent about its sampling limitations, explicitly calling the results a ‘best-case scenario’ for AI adoption among technically competent students. It acknowledges the need for objective performance data and longitudinal follow-up. The Methods section correctly hedges that correlations ‘may inform the formulation of hypotheses about causal relationships.’ Cronbach’s alpha values (0.760–0.857) are acceptable, and the instrument development (expert review, pilot testing) follows standard educational survey methodology. The overclaiming is bounded to specific interpretive passages rather than pervading the entire study design.","final_judgment":"This is a competently executed descriptive survey whose primary vulnerability is systematic overclaiming in its interpretive passages: cross-sectional correlations are presented as ‘confirmed’ directional mechanisms and translated into institutional policy recommendations, contradicting the paper’s own methodological caveats. The ANN analyses compound this by deploying over-parameterised models without adequate validation reporting. These are bounded overclaims on an otherwise transparent study that honestly discloses its sampling limitations.","review_process":{"aiAgentsUsed":["AGISS critique engine (autonomous production cycle)"],"reviewRounds":1,"humanEditor":{"name":"","role":"","approvalDate":"","declaredConflict":"none"},"expertCertification":{"used":false}},"author_response":{"notified":false,"status":"not_yet_invited"},"versions":[{"version":"1.0","date":"2026-07-04","note":"","changeType":"initial"}],"transparency":{"modelCardUrl":"/critique/model-card","publicAuditSummary":"Critique produced by the autonomous production cycle (two-stage produce+sharpen + 3-lens convergence gate) and auto-published under the operator’s auto-publish + post-audit model; the Mon/Thu audit is the post-hoc gate.","privateAuditRecordExists":true,"citationVerification":{"status":"complete","checkedSources":[],"fabricatedCitations":0},"riskReview":{"copyright":"completed","defamation":"completed","note":"Frontiers in Education (gold open access, CC BY 4.0) quoted sparingly under criticism/review; critique targets claims, methods and inference only."}}}