Expanding the Good, Exposing the Weak: A Critical Analysis of Generative Artificial Intelligence and Relational Trust in Latin American Higher Education A Critical Evaluation of Originality, Methodological Rigour, Statistical Validity, Theoretical Contribution, and Institutional Implications

 

Expanding the Good, Exposing the Weak: A Critical Analysis of Generative Artificial Intelligence and Relational Trust in Latin American Higher Education

A Critical Evaluation of Originality, Methodological Rigour, Statistical Validity, Theoretical Contribution, and Institutional Implications

Critical Review of the Original Research Article

Original Journal Name: Interpersona: An International Journal on Personal Relationships
Original Article Title: Generative Artificial Intelligence and Relational Trust in Latin American Higher Education
Authors of the Original Research Paper: Castro Valencia, Soto Pérez, and Rivera Ramírez
Journal Volume: Volume 20
Issue: Issue 4
Year of Publication: 2026
Original Article DOI: 10.54899/ijpr.v20i4.1023

Critical Analysis Prepared By

Dr. Mamta Vyas
Independent Researcher
Management and Economics 

Abstract

Generative artificial intelligence has rapidly transformed higher education, creating new opportunities for personalised learning, academic support, and institutional efficiency while simultaneously raising concerns about trust, fairness, academic integrity, and human interaction. The research paper “Generative Artificial Intelligence and Relational Trust in Latin American Higher Education” by Castro Valencia, Soto Pérez, and Rivera Ramírez offers an important conceptual contribution by framing AI adoption as a relational transformation rather than merely a technological or pedagogical development. This critical analytical study evaluates the paper's originality, significance, methodology, statistical validity, theoretical integration, and practical implications. The original research reports that 92% of students and 79% of lecturers engage with generative AI, identifies a 31-percentage-point difference between student support for AI-assisted feedback and lecturer provision, and finds that economic development and internet penetration jointly explain approximately 60.4% of the variance in national AI readiness across 13 Latin American countries. The critical analysis identifies substantial strengths in the paper's interdisciplinary framing, regional relevance, use of secondary evidence, and recognition of structural inequality. However, important limitations remain concerning the operationalisation of relational trust, comparability of survey indicators, integration of micro-level and macro-level evidence, statistical uncertainty, and interpretation of cross-sectional regression results. The study concludes that the original article offers a promising theoretical foundation but requires stronger construct measurement, empirical integration, robustness analysis, and more cautious interpretation to establish the relational divide as an empirically validated phenomenon.

Keywords: Generative Artificial Intelligence, Relational Trust, Higher Education, Critical Research Analysis, Digital Divide, Academic Integrity.

 INTRODUCTION

1.1 Background of the Study

Generative artificial intelligence has emerged as one of the most significant technological developments affecting contemporary higher education.

AI applications can support teaching, learning, academic writing, research, assessment, student guidance, and administrative processes.

However, the expansion of AI creates important institutional and interpersonal questions.

Universities are not merely technological environments. They are social institutions in which academic authority, student development, professional trust, fairness, and interpersonal relationships play essential roles.

The research paper by Castro Valencia et al. [1] attempts to address this issue by arguing that AI adoption should be understood as a transformation of the relational fabric of higher education.

The authors introduce the idea of a relational divide, suggesting that the consequences of AI adoption may differ according to the structural resources available to countries and institutions.

This critical review evaluates whether the empirical evidence presented in the original paper adequately supports this proposition.

LITERATURE REVIEW

.Interpersonal Trust and Academic Relationships

Mayer, Davis, and Schoorman [6] developed an influential integrative model of organisational trust based on ability, benevolence, and integrity.

Their framework provides a strong theoretical foundation for examining trust in educational relationships.

However, the original research paper does not directly map its survey indicators onto these dimensions.

Rousseau et al. [7] further explain that trust is a cross-disciplinary concept associated with accepted vulnerability and positive expectations.

This suggests that trust should be measured through perceptions of a specific party rather than inferred exclusively from general satisfaction or behavioural preferences.

 AI Adoption in Higher Education

Davis [8] developed the Technology Acceptance Model, highlighting perceived usefulness and perceived ease of use as important determinants of technology adoption.

Venkatesh et al. [9] extended this approach through the Unified Theory of Acceptance and Use of Technology.

These models are relevant to AI adoption, but they do not independently explain the interpersonal consequences of AI-mediated educational interaction.

The original article correctly identifies this theoretical distinction.

However, a stronger empirical analysis would integrate technology acceptance and interpersonal trust as separate but potentially related constructs.

 Academic Integrity and AI Governance

Cotton, Cotton, and Shipway [10] examine the implications of ChatGPT for academic integrity.

UNESCO [11] advocates a human-centred approach to generative AI in education and research.

These sources demonstrate that AI governance is not simply a question of technological efficiency. It also involves fairness, transparency, institutional responsibility, and the protection of educational values.

The original article's relational perspective is therefore relevant to the broader governance debate.

 Digital Inequality and Structural Readiness

Warschauer [12] and van Dijk [13] explain that digital inequality extends beyond physical access to technology.

It also includes skills, autonomy, and the capacity to derive meaningful benefits from digital resources.

This literature provides a useful foundation for the relational-divide concept.

However, the original article must demonstrate more directly how digital and structural inequalities translate into specific interpersonal outcomes.

The existing literature supports the plausibility of this relationship, but plausibility should not be confused with empirical confirmation.

 Research Gap

The central research gap identified through this critical review is the absence of a sufficiently integrated empirical model linking structural AI readiness with directly measured interpersonal trust in higher education.

The original paper offers a conceptual bridge between these domains, but future studies must validate the bridge through more precise measurement and multilevel analysis.

1.2 Statement of the Research Problem

The primary research problem is the potential mismatch between the theoretical importance assigned to relational trust and the empirical indicators used to measure it.

The original paper discusses trust, fairness, social cohesion, and interpersonal relationships extensively.

However, its empirical analysis relies largely on aggregate percentages and a country-level regression involving economic development, internet penetration, and national AI readiness.

The central question is whether these measures adequately demonstrate that generative AI is creating a relational divide.

The problem can be expressed as follows:

Theoretical ambition exceeds empirical operationalisation.

This does not invalidate the research. Rather, it identifies the principal area in which further analytical development is required.

1.3 Research Objectives

The present critical review aims to:

1. Evaluate the originality and significance of the original article.
2. Examine the adequacy of its research design and data sources.
3. Assess the construct validity of relational trust indicators.
4. Critically interpret the reported statistical findings.
5. Identify weaknesses in theoretical integration and empirical reasoning.
6. Evaluate the paper's relevance for higher education institutions and AI governance.
7. Develop recommendations for strengthening future research.

1.4 Research Questions

RQ1: To what extent does the original paper demonstrate originality in conceptualising AI adoption as a relational phenomenon?

RQ2: Are the research design, variables, and statistical methods adequate to support the proposed relational-divide concept?

RQ3: Does the empirical evidence adequately distinguish relational trust from technological acceptance, fairness, and institutional satisfaction?

RQ4: How should the reported statistical results be interpreted in view of the small cross-national sample?

RQ5: What methodological and theoretical improvements could strengthen the original research?

Significance of the Critical Review

The review is relevant to researchers, higher education administrators, lecturers, policymakers, and scholars of management and educational technology.

It provides a structured understanding of the difference between a theoretically compelling research paper and an empirically validated contribution.

The analysis also demonstrates the importance of evaluating research claims through the alignment of theory, measurement, evidence, and statistical interpretation.

2. METHODOLOGY

2.1 Research Design

The present study adopts a critical analytical review methodology to evaluate the research paper titled “Generative Artificial Intelligence and Relational Trust in Latin American Higher Education” by Castro Valencia, Soto Pérez, and Rivera Ramírez, published in Interpersona in 2026 [1].

The purpose of this critical analysis is not merely to summarise the original article but to systematically examine the relationship between its theoretical propositions, empirical evidence, statistical techniques, methodological choices, and conclusions.

The analysis follows a structured evaluative framework encompassing seven major dimensions:

1. Originality and theoretical contribution.
2. Significance and relevance of the research problem.
3. Methodological rigour and research design.
4. Statistical correctness and empirical validity.
5. Literature review and theoretical integration.
6. Clarity, presentation, and scholarly communication.
7. Practical implications and future research potential.

The original article is evaluated against the expectations of rigorous academic research, particularly the alignment between research objectives, measurable constructs, data sources, analytical procedures, and conclusions.

The critical review also considers the supplied journal-preparation requirements, including title-page specifications, abstract length, methodological transparency, results and discussion, conclusion, literature review, and IEEE reference formatting.

2.2 Data Sources

The primary source of this critical analysis is the complete 32-page original research article published in Interpersona [1].

The original article reports using the following secondary data sources:

• Digital Education Council Artificial Intelligence in Higher Education Latin America Survey 2026.
• Latin American Artificial Intelligence Index (ILIA 2024).
• International Monetary Fund World Economic Outlook database.
• International Telecommunication Union digital development statistics.
• Oxford Insights Government AI Readiness Index.
• Earlier empirical and theoretical literature on trust, technology acceptance, academic integrity, and educational relationships.

The critical review relies on the data, tables, statistical results, and interpretations reported in the original article. No new respondent-level dataset has been collected for this paper.

2.3 Analytical Techniques

The study employs qualitative critical evaluation, comparative interpretation, construct-validity assessment, statistical-reporting review, and methodological appraisal.

The reported numerical findings are examined for their appropriateness and interpretive limits.

Particular attention is given to:

• The 31-percentage-point difference between student support for AI-assisted feedback and lecturer provision of AI-assisted feedback.
• The reported regression model involving 13 Latin American countries.
• The relationship between economic development, internet penetration, and national AI readiness.
• The distinction between relational trust and proxy indicators.
• The integration of individual-level survey evidence and country-level structural indicators.

The critical review does not independently reproduce the original regression because the underlying country-level dataset, survey microdata, and complete computational code are not supplied as part of the material reviewed.

2.4 Ethical Considerations

This study is based on an already published research article and its cited secondary sources. It does not involve new human participants, interviews, experiments, or personal data.

The analysis seeks to maintain academic fairness by distinguishing between demonstrated weaknesses, plausible limitations, and areas requiring further verification.

The authors of the original paper are not evaluated personally. The assessment focuses exclusively on the research design, evidence, analytical procedures, and scholarly contribution.

 3. RESULTS AND DISCUSSION

3.1 Overall Assessment of the Original Research Paper

The original paper examines an increasingly important question: whether generative AI changes not only the manner in which students and lecturers perform academic tasks but also the interpersonal relationships through which educational institutions function.

Its central argument is that AI adoption may generate a “relational divide” alongside the conventional digital divide. This concept suggests that unequal infrastructure, governance, institutional resources, and AI literacy may influence the extent to which educational communities can integrate AI without damaging trust, fairness, and social cohesion.

The conceptual contribution is significant because conventional AI-in-education research frequently emphasises productivity, learning outcomes, technological acceptance, or academic misconduct. The original article instead considers the social and relational consequences of AI-mediated academic practices.

Nevertheless, the strength of a theoretical proposition depends on the quality of the evidence supporting it.

The principal critical observation is that the original paper develops an intellectually ambitious theoretical framework but provides comparatively limited direct empirical measurement of its central construct: relational trust.

The article is therefore best understood as an exploratory and conceptually oriented empirical study rather than a definitive validation of the relational divide.

3.2 Critical Evaluation of Originality

3.2.1 Strengths of the Originality

The paper's strongest original contribution is its reframing of generative AI as a relational technology.

The authors argue that AI systems increasingly perform tasks traditionally embedded in interpersonal academic relationships, including:

• Providing academic feedback.
• Explaining difficult concepts.
• Evaluating assignments.
• Supporting student guidance.
• Assisting with communication.
• Influencing perceptions of academic effort and authenticity.

This framing extends conventional technology-acceptance approaches by recognising that technological adoption can alter interpersonal expectations.

The introduction of the relational-divide concept is particularly noteworthy. The concept connects structural inequalities in AI infrastructure and governance with the potential quality of student–teacher and student–student relationships.

The regional focus on Latin America also strengthens the paper's originality. Research on AI in higher education has frequently concentrated on North American, European, and technologically advanced institutional settings. The paper contributes a perspective from a region characterised by considerable differences in national AI readiness.

3.2.2 Weaknesses in the Originality Claim

Although the conceptual framing is promising, the empirical findings are not entirely novel.

High AI adoption among students, concerns about academic integrity, lecturer caution regarding assessment, and unequal technological readiness have already been discussed in the literature [2]–[5].

The paper's novelty therefore rests primarily on the integration of these issues within a relational framework rather than on the discovery of entirely new empirical patterns.

The authors also do not fully establish whether the relational divide represents a genuinely new theoretical construct or an extension of existing concepts such as:

• Digital inequality.
• Social capital.
• Institutional trust.
• Educational inequality.
• Technology-mediated social presence.
• AI governance inequality.

A stronger originality claim would require a precise explanation of what the relational divide captures that these established concepts cannot adequately explain.

Critical judgement: The paper demonstrates moderate-to-high conceptual originality, but its empirical originality is more limited.

3.3 Critical Analysis of the Research Problem and Significance

The research problem is timely and relevant.

Universities increasingly face a difficult institutional choice between encouraging AI adoption and preserving the human relationships that underpin academic learning.

The original paper correctly recognises that AI can simultaneously produce educational benefits and relational concerns.

Students may appreciate rapid feedback, personalised explanations, and greater academic flexibility. Lecturers may value efficiency and assistance with routine work while remaining concerned about assessment authenticity, professional judgement, and the quality of human interaction.

The research is therefore significant for three reasons.

First, it contributes to the understanding of academic relationships in technology-mediated environments.

Second, it highlights the possibility that AI adoption may be unevenly distributed across institutions and countries.

Third, it offers implications for AI governance, educational policy, academic integrity, and institutional management.

However, the significance of the paper is reduced by the difference between its broad theoretical ambition and its empirical scope.

The article seeks to explain how structural inequalities influence relational experiences, yet the empirical analysis does not directly connect country-level AI readiness with individual-level relational trust.

Consequently, the paper establishes the importance of the problem more convincingly than it establishes the proposed mechanism through which the problem operates.

3.4 Critical Evaluation of Research Methodology

3.4.1 Research Design

The authors employ a quantitative secondary-data design described as having convergent logic.

The design combines two analytical strands:

1. A relational analysis of student and lecturer attitudes.
2. A structural analysis of national AI readiness across 13 Latin American countries.

The use of secondary data is appropriate for exploratory regional research, particularly where large-scale survey information is available.

The authors also appropriately acknowledge that their cross-sectional correlational design does not establish causal relationships.

However, the convergent design is insufficiently integrated.

The relational analysis and structural regression are presented as complementary sources of evidence, but the study does not estimate a unified model connecting national readiness with institutional or individual relational outcomes.

This creates a methodological gap.

A genuinely integrated analytical design could have linked institutional or national identifiers in the survey dataset to country-level readiness indicators, subject to data access and appropriate safeguards.

For example, the following relationship could be examined:

National AI readiness → Institutional AI governance → Lecturer AI literacy → Student perceptions of fairness → Relational trust.

Such a framework would provide a clearer empirical explanation of the proposed relational divide.

3.4.2 Secondary Data Transparency

The original paper reports more than 30,000 survey responses across 29 higher education institutions.

This is a considerable empirical resource.

However, the paper does not provide sufficiently detailed information concerning:

• Sampling procedures.
• Institutional selection criteria.
• Survey response rates.
• Exact questionnaire wording.
• Sampling weights.
• Reliability of individual survey items.
• Representativeness of the respondent population.
• The extent to which the survey covers different institutional types and countries.

These omissions limit the reader's ability to evaluate the representativeness and reliability of the survey estimates.

A large respondent count does not automatically ensure that a sample represents the broader higher education population.

If the institutions participating in the survey were disproportionately technology-oriented, the reported adoption rates might not accurately represent all universities in Latin America.

Critical judgement: The data sources appear relevant, but the transparency of their sampling and measurement procedures is insufficient for strong regional generalisation.

3.5 Critical Analysis of Construct Validity

Construct validity is one of the most important weaknesses identified in the original research.

The paper extensively invokes interpersonal trust theory, particularly the framework developed by Mayer, Davis, and Schoorman [6].

According to this framework, trust is associated with perceptions of:

• Ability.
• Benevolence.
• Integrity.

However, the original paper operationalises relational orientations through indicators such as:

• Fear of peer misuse of AI.
• Student support for AI-assisted feedback.
• Institutional AI satisfaction.
• Lecturer AI feedback practices.
• Perceptions of AI as a threat to professional roles.

These indicators are relevant to the broader relational environment, but they do not directly measure the three dimensions of interpersonal trust.

For instance, fear that peers misuse AI may indicate concerns about academic fairness. It does not necessarily establish that students distrust their peers in a general interpersonal sense.

Similarly, support for AI-assisted feedback may indicate perceived usefulness, convenience, or technological acceptance rather than trust in the lecturer or the AI system.

The distinction is important because trust, acceptance, satisfaction, fairness, and perceived usefulness are related but theoretically different constructs.

A stronger empirical model would distinguish them.

Construct	What it measures	Potential indicator
Relational trust	Confidence in another party's conduct	Perceived ability, benevolence, integrity
AI acceptance	Willingness to use AI	Adoption intention and actual use
Perceived fairness	Evaluation of equitable treatment	Fairness of AI-assisted assessment
Institutional satisfaction	Evaluation of institutional policy	Satisfaction with AI guidelines
Academic integrity concern	Concern regarding authentic work	Perceived misuse by peers
Social presence	Perceived human connection	Feeling recognised and supported

The original paper would benefit from validating these constructs separately before combining them into a relational interpretation.

3.6 Critical Analysis of the Human–Feedback Paradox

The paper identifies a 31-percentage-point difference between student support for AI-assisted feedback and lecturer provision of AI-assisted feedback.

The reported figures are:

Indicator	Percentage
Students supporting lecturers' use of AI for feedback	50%
Lecturers providing AI-assisted feedback	19%
Reported percentage-point difference	31

The finding is engaging and potentially valuable.

It suggests that student expectations regarding AI-mediated feedback may differ from current lecturer practices.

However, the interpretation requires caution.

The two indicators measure different dimensions of behaviour.

The student indicator measures an attitude or preference.

The lecturer indicator measures reported practice.

Therefore, the 31-percentage-point difference is not a direct measure of a single behavioural gap.

It is a descriptive comparison between two aggregated percentages.

The difference could arise from several factors:

• Students' preference for faster feedback.
• Lecturers' concerns about accuracy.
• Institutional restrictions.
• Lack of AI training.
• Professional responsibility.
• Privacy concerns.
• Assessment-related liability.
• Different expectations regarding academic quality.

The original authors interpret the gap primarily as a relational boundary protecting human judgement.

This explanation is plausible, but it is not uniquely established by the evidence.

The alternative explanations require investigation.

A more rigorous study would examine whether lecturer decisions are explained by relational concerns, technological limitations, institutional governance, or a combination of these factors.

The distinction has practical importance. If the gap is primarily due to inadequate training, training interventions may reduce it. If the gap is based on professional and ethical concerns, training alone may not change lecturer behaviour.

3.7 Critical Statistical Analysis of National AI Readiness

The original paper estimates ordinary least squares regression models using data from 13 Latin American countries.

The dependent variable is the national AI readiness score.

The independent variables are:

• Logarithm of GDP per capita at purchasing power parity.
• Internet penetration.

The reported findings are summarised below.

Table 1. Reported Regression Model Results

Statistical measure	Model 1: Income	Model 2: Internet	Model 3: Combined
R²	0.531	0.512	0.604
Adjusted R²	0.489	0.468	0.525
Sample size	13	13	13
Overall F-statistic	9.96	5.07	12.83
Overall model significance	p < 0.01	p < 0.05	p < 0.01

Source: Original article [1].

Interpretation

The combined model reports an R² of 0.604.

This indicates that the two predictors jointly account for approximately 60.4% of the observed cross-country variation in the AI readiness score within the sample.

The adjusted R² is 0.525, indicating that after accounting for the number of predictors and the small sample size, the model explains approximately 52.5% of the variation.

The distinction between R² and adjusted R² is important.

The original paper reports both measures, which is a methodological strength.

Nevertheless, several concerns remain.

3.7.1 Small Sample Size

The regression contains 13 country observations and two predictors.

Although the authors use HC3 heteroscedasticity-consistent standard errors, the small sample remains a serious limitation.

Robust standard errors can address certain forms of heteroscedasticity but do not eliminate the limitations associated with low statistical power, unstable coefficients, or influential observations.

The regression results may be sensitive to the inclusion or exclusion of individual countries.

3.7.2 Multicollinearity

The paper reports a correlation of 0.73 between income and internet penetration.

This indicates that the predictors are substantially related.

The reported variance inflation factor is 2.13, which does not indicate severe multicollinearity under commonly used thresholds.

However, the overlap between economic development and internet penetration complicates the interpretation of their individual coefficients.

The income coefficient becomes marginally significant in the combined model, while internet penetration remains statistically significant at the reported 5% level.

This suggests that internet access may contain additional explanatory information beyond income in the sample.

It does not, however, establish that internet penetration independently causes higher AI readiness.

3.7.3 Missing Robustness Checks

The original study would be strengthened by conducting:

• Leave-one-country-out regression analysis.
• Cook's distance and leverage diagnostics.
• Alternative regression specifications.
• Sensitivity analysis using alternative readiness measures.
• Confidence intervals for regression coefficients.
• Examination of influential countries such as Brazil, Chile, and Mexico.

These checks would help establish whether the reported 60.4% explanatory power is stable.

3.7.4 Interpretation of Statistical Significance

The reported overall model significance is p = 0.002.

This provides evidence against the null hypothesis that all slope coefficients are zero under the model assumptions.

However, statistical significance should not be confused with theoretical validation.

The regression supports an association between national economic and digital conditions and AI readiness.

It does not directly test whether these conditions produce greater relational trust or prevent relational harm.

Critical judgement: The statistical analysis is informative but should be interpreted as exploratory evidence of structural stratification rather than a definitive empirical test of the relational divide.

3.8 Critical Evaluation of the Theoretical Framework

The original paper draws upon several major theoretical traditions:

• Interpersonal trust theory.
• Social capital theory.
• Relational sociology.
• Technology acceptance theory.
• Social presence theory.
• Self-determination theory.
• Digital divide theory.

The interdisciplinary approach is a major strength.

However, the framework is comparatively broad.

Several theories are extensively discussed but are not directly translated into testable variables or analytical models.

For example, social capital theory is used to explain the importance of relationships in academic communities, but the empirical analysis does not measure social capital.

Similarly, trust theory is invoked to explain the relational consequences of AI feedback, but the study does not directly test trust dimensions.

The problem is not that theoretical diversity is undesirable. The problem is that theoretical breadth should be matched by empirical depth.

A more focused theoretical framework could centre on three constructs:

1. Interpersonal trust.
2. AI-mediated academic interaction.
3. Institutional governance.

This would allow the authors to develop clearer hypotheses and testable mechanisms.

3.9 Critical Evaluation of Results and Discussion

The original article presents its descriptive findings clearly.

The reported adoption rates and feedback comparisons are easy to understand.

The tables and figures also improve the presentation of regional and cross-national patterns.

However, some interpretations are stronger than the available evidence.

For example, the article argues that adoption has outpaced institutional policy.

Although this interpretation is plausible, the reported survey percentages do not independently measure the chronological relationship between adoption and institutional policy development.

Similarly, the article interprets the lecturer feedback gap as evidence of deliberate relational boundary-setting.

The survey figures alone cannot establish the underlying motivations of lecturers.

A more rigorous discussion would distinguish between:

• Direct empirical findings.
• Theoretical interpretations.
• Plausible alternative explanations.
• Claims requiring further research.

Such distinctions would improve the paper's analytical precision.

3.10 Critical Evaluation of Writing and Presentation

The original paper is generally well-written and intellectually engaging.

The authors employ a sophisticated academic style and present the theoretical argument coherently.

The concept of relational trust provides a clear thematic focus.

However, the article occasionally uses strong rhetorical formulations that exceed the empirical evidence.

For example, the argument that the most consequential effects of AI may concern interpersonal relationships rather than learning outcomes is an important proposition, but it is not directly tested through comparative learning-outcome data.

The article also follows a conventional academic structure with an abstract, introduction, theoretical framework, methodology, results and discussion, conclusion, and references.

This is appropriate for the published journal article.

However, the journal-preparation guidelines supplied for the present critical review specify a different order of sections. The critical review therefore follows those supplied guidelines rather than reproducing the original article's sequence.

A further distinction is necessary: the original article does contain a conclusion section. The central criticism is not that the conclusion is missing, but that the conclusion's strongest theoretical claims require more direct empirical validation.

CONCLUSION

4.1 Summary of Key Findings

The critical analysis of “Generative Artificial Intelligence and Relational Trust in Latin American Higher Education” establishes that the original article is conceptually ambitious, timely, and relevant to the changing landscape of higher education.

Its most important contribution is the attempt to move beyond a narrow technological understanding of AI adoption and examine the implications of AI for academic relationships.

The concept of the relational divide is potentially valuable because it connects national and institutional inequality with the quality of AI-mediated academic interaction.

However, the review identifies several important weaknesses.

First, relational trust is theoretically discussed but not directly measured through validated trust constructs.

Second, the 31-percentage-point feedback difference compares student attitudes with lecturer practices, making it an informative descriptive contrast but not a direct measure of a unified behavioural gap.

Third, the structural regression uses only 13 country observations, limiting statistical power and the reliability of broader generalisations.

Fourth, the two analytical strands are not integrated into a unified multilevel model.

Fifth, alternative explanations for lecturer behaviour are insufficiently examined.

These weaknesses reduce the strength of the empirical validation but do not eliminate the value of the original conceptual contribution.

4.2 Implications for Theory

The original article provides an opportunity to extend interpersonal trust theory into AI-mediated educational settings.

However, the theoretical extension should be accompanied by clearer operational definitions.

Future studies should distinguish between trust in AI systems, trust in lecturers, trust in peers, and trust in institutional governance.

The relational-divide concept should also be positioned more precisely in relation to established theories of digital inequality and social capital.

A stronger conceptual model could demonstrate how structural conditions influence institutional governance, which subsequently affects interpersonal trust and perceived fairness.

4.3 Implications for Practice

Higher education institutions should avoid treating AI adoption as a purely technological decision.

Institutional AI policy should address:

• Transparent disclosure of AI use.
• Fair assessment practices.
• Lecturer training.
• Student AI literacy.
• Human oversight.
• Privacy and data protection.
• Academic integrity.
• Accessibility and equitable infrastructure.

AI-assisted feedback should not automatically be considered a replacement for human academic judgement.

Instead, AI may be used to support formative feedback, while lecturers retain responsibility for evaluative decisions, personal guidance, and the communication of academic expectations.

The original article's conceptual framing is useful for developing such policies, even though its empirical conclusions require further validation.

4.4 Limitations of the Present Critical Review

This critical review is based on the published article and the secondary evidence reported by its authors.

The review does not independently access the original survey microdata or reproduce the regression models.

Therefore, it evaluates the adequacy of the reported methodology and statistical interpretation rather than independently verifying every underlying numerical estimate.

The review also does not constitute a systematic review of the entire AI-in-higher-education literature.

4.5 Recommendations for Future Research

Future research should pursue the following directions.

First, direct measurement of relational trust. Validated instruments should be used to measure ability, benevolence, integrity, and interpersonal trust.

Second, multilevel research design. Student-level, lecturer-level, institutional-level, and country-level data should be linked where appropriate.

Third, longitudinal analysis. Repeated observations should examine whether changes in AI governance are associated with changes in academic trust.

Fourth, qualitative investigation. Interviews and focus groups could reveal why lecturers and students differ in their attitudes towards AI-mediated feedback.

Fifth, robust statistical testing. Larger cross-national samples and sensitivity analysis should be used to establish whether the structural relationships are stable.

Sixth, comparative research. The relational-divide proposition should be examined across Latin America, Asia, Africa, Europe, and other educational contexts.

Seventh, institutional policy evaluation. Future research should assess whether transparent AI governance, training, and human oversight improve trust and fairness.

Final Critical Judgement

The original research paper is a promising contribution to the interdisciplinary study of generative AI and higher education. Its principal strength lies in its conceptual reframing of AI adoption as a relational transformation. Its principal weakness lies in the limited empirical operationalisation of the central construct and the incomplete integration of its analytical strands.

The paper should therefore be regarded as a valuable exploratory study that opens an important avenue for future research rather than as conclusive empirical proof of the relational divide.

The most significant lesson is that strong academic research requires more than an original idea. It requires a precise construct, a transparent methodology, reliable measurement, appropriate statistical analysis, and conclusions proportionate to the evidence.

The contribution of critical analysis is precisely this: to expand what is valuable in the original research while exposing the weaknesses that must be addressed for the work to become more rigorous, reproducible, and influential.

REFERENCES

IEEE numbered citation style is used below. The references are selected from the original paper's bibliography and correspond to the citations used in this critical review.

[1] A. M. C. Valencia, M. S. Pérez, and J. C. R. Ramírez, “Generative artificial intelligence and relational trust in Latin American higher education,” Interpersona, vol. 20, no. 4, 2026. doi: 10.54899/ijpr.v20i4.1023.

[2] C. K. Y. Chan and W. Hu, “Students’ voices on generative AI: Perceptions, benefits, and challenges in higher education,” International Journal of Educational Technology in Higher Education, vol. 20, no. 1, Art. no. 43, 2023. doi: 10.1186/s41239-023-00411-8.

[3] C. Stöhr, A. W. Ou, and H. Malmström, “Perceptions and usage of AI chatbots among students in higher education across genders, academic levels and fields of study,” Computers and Education: Artificial Intelligence, vol. 7, Art. no. 100259, 2024. doi: 10.1016/j.caeai.2024.100259.

[4] A. Kasneci et al., “ChatGPT for good? On opportunities and challenges of large language models for education,” Learning and Individual Differences, vol. 103, Art. no. 102274, 2023. doi: 10.1016/j.lindif.2023.102274.

[5] S. Z. Salas-Pilco and Y. Yang, “Artificial intelligence applications in Latin American higher education: A systematic review,” International Journal of Educational Technology in Higher Education, vol. 19, no. 1, Art. no. 21, 2022. doi: 10.1186/s41239-022-00326-w.

[6] R. C. Mayer, J. H. Davis, and F. D. Schoorman, “An integrative model of organizational trust,” Academy of Management Review, vol. 20, no. 3, pp. 709–734, 1995. doi: 10.5465/amr.1995.9508080335.

[7] D. M. Rousseau, S. B. Sitkin, R. S. Burt, and C. Camerer, “Not so different after all: A cross-discipline view of trust,” Academy of Management Review, vol. 23, no. 3, pp. 393–404, 1998. doi: 10.5465/amr.1998.926617.

[8] F. D. Davis, “Perceived usefulness, perceived ease of use, and user acceptance of information technology,” MIS Quarterly, vol. 13, no. 3, pp. 319–340, 1989. doi: 10.2307/249008.

[9] V. Venkatesh, M. G. Morris, G. B. Davis, and F. D. Davis, “User acceptance of information technology: Toward a unified view,” MIS Quarterly, vol. 27, no. 3, pp. 425–478, 2003. doi: 10.2307/30036540.

[10] D. R. E. Cotton, P. A. Cotton, and J. R. Shipway, “Chatting and cheating: Ensuring academic integrity in the era of ChatGPT,” Innovations in Education and Teaching International, vol. 61, no. 2, pp. 228–239, 2024. doi: 10.1080/14703297.2023.2190148.

[11] UNESCO, Guidance for Generative AI in Education and Research. Paris, France: UNESCO, 2023. doi: 10.54675/EWZM9535.

[12] M. Warschauer, Technology and Social Inclusion: Rethinking the Digital Divide. Cambridge, MA, USA: MIT Press, 2003.

[13] J. A. G. M. van Dijk, The Digital Divide. Cambridge, UK: Polity Press, 2020.