Exploring the Determinants of the Sustainable Use of Artificial Intelligence in Peruvian University Teachers: A Structural Equation Modeling Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a useful and professionally conducted investigation on a timely subject.
 
Major observations
(A) I would kindly suggest including age and gender in the SEM model as independent predictors for teaching concerns (TC) and perceived ethics (PE). Two reasons would support this approach: (a1) for testing moderation on the Outcome, the model should comprise Predictor, Moderator, and Predictor x Moderator; (a2) the resulting SEM model might be better. If such a model has been tested and discarded, a supplementary material should describe it (i.e., supplementary material should be provided for interested readers).
 
(B) I strongly recommend having the concepts, context, and rationale described with appropriate references in the first part of introduction (as they partly are in the present form), followed by objective and specific aims of this investigation free of citations/references.
 
(C) Title of section 2 should be changed. For example, a phrase such as "Research constructs and hypotheses" might accommodate the content.
 
(D) In the present form, sections of methods and results are mixed. They should be clearly separated, and methods must be clear and comprehensive. This issue is two-fold: (d1) all evaluation criteria and thresholds, together with appropriate references/citations must be specified in the section of methods; (d2) results must comprise bare findings and be reference free.
 
(E) Section of limitations specifies the cross-sectional design and its subsequent limitations. I kindly recommend specifying the design in the abstract, too. I would also recommend including two additional limitations related to the design: (e1) causal relationships cannot be established; (e2) it only captured relationships within a limited time window, while the research investigated a highly dynamic field and subsequent relationships.
 
Additional observations
(a) I recommend introducing and subsequent use of acronym AI in the abstract. The abstract might also include some information about design (already mentioned at point E above), anonymous character of the questionnaire, demographics regarding the respondents, and essential numerical/quantitative findings.

(b) I kindly recommend having an alphabetical list of acronyms at the end of the manuscript.

(c) The management of references must be revised (the style of references and citations do not comply with the required standards + some essential information is missing from the references).

(d) Throughout the manuscript, many sentences begin with a citation. This should be corrected; for example, sentences could begin with the name of the first author followed by the citation.

(e) For better conveying the message in section 2.2, I recommend improving the visual <connection> of each hypothesis with the rationale behind (i.e., the text that precedes the hypothesis).

(f) In Figures 1 and 2, I recommend having the variables "age" and "gender" in small letters (lowercase), in contrast to the constructs' acronyms.

(g) Figure 1 would benefit from having the hypotheses (e.g. H1, H2, etc.) as labels on the arrows.

(h) I would recommend moving Table 2 into supplementary material. Moreover, additional horizontal lines would help better reading.

(i) All legends/captions of figures and tables should be more informative (figures and tables should be self-explanatory). In addition, the acronyms should be explained in the footnote of each table.

(j) You should avoid having p-values specified as 0.000. Specify them as <0.001 in tables, as well.

(k) Within the text, you should choose between expressing statistical significance either by stars or p-values (not using both). I would opt for p-values, rather than stars in the text/narrative.

 
Factual errors to be corrected
(1) Title of section 3 should be "3. Methods" (i.e., a plural form).

(2) Double-check Tables 1 and 4 – everything should be in English.

(3) Reference "Anderson and Kim (2024)" is missing.

(4) The threshold of SRMR CANNOT be 0.85. While the actual threshold value is a matter of choice and discussion, it should be one order of magnitude lower than the value you specified (for example, 0.08 or 0.1).

(5) In Table 5, standard deviations of regression coefficients are expressed, but the table's footnote mentions SE (probably the standard error). I would kindly recommend having the standard errors in the table (they are more suitable, as they are used in the confidence intervals).

Author Response

REVIEWER 1

Main Comments

(A) I kindly suggest including age and gender as independent predictors in the SEM model for teaching concerns (TC) and perceived ethics (PE). Two reasons support this approach:
(a1) To test moderation in the outcome, the model should include Predictor, Moderator, and Predictor × Moderator.
(a2) The resulting SEM model could be improved. If such a model has been tested and discarded, it should be described in supplementary material (i.e., additional material should be provided for interested readers).

Response: Thank you for your suggestion. However, these changes cannot be made for the following reasons:

• The relationships established in the SEM model (including moderation hypotheses) are supported by the literature.
• The proposed model, in addition to altering the findings, does not fit the data well, as this analysis was conducted in SmartPLS. Moreover, this information cannot be provided because the analysis results were not saved, and due to software licensing restrictions, we cannot rerun the analysis.

(B) I strongly recommend describing the concepts, context, and justification with appropriate references in the first part of the introduction (as they are partially included in the current version), followed by the specific objectives and goals of this research without citations/references.

Response: Thank you for your comment. Changes have been made in the introduction as follows:

• It now begins with conceptual background before introducing specific constructs.
• The theoretical framework is separated from the research objectives.
• A clear context and justification are provided.
• A logical flow is maintained, progressing from general concepts to a specific research focus.

(C) The title of Section 2 should be changed. For example, a phrase such as “Constructs and Research Hypotheses” could better reflect the content.

Response: Thank you for your suggestion. The necessary changes have been made.

(D) In its current form, the methods and results sections are intermingled. They should be clearly separated, and the methods should be detailed and complete. This issue has two parts:
(d1) All evaluation criteria and thresholds, along with appropriate references/citations, should be specified in the methods section.
(d2) The results should consist of findings only and be free of references.

Response: Thank you for your recommendation. However, it is important to clarify that methods and results (as with all sections of an article) are complementary. In this case, the goal is to present the findings clearly and avoid confusing potential readers. While the methodology section describes the procedure used in data analysis, the results section presents the findings. As is customary in the scientific community, citations are included to reference the decision thresholds established by previous authors. Therefore, to enhance readability and prevent confusion, it is necessary to maintain the presentation of results along with citations to the authors of the decision thresholds. Thank you.

(E) The limitations section specifies the cross-sectional design and its associated constraints. I kindly recommend specifying the study design in the abstract as well. Additionally, I suggest including two additional limitations related to the design:
(e1) Causal relationships cannot be established.
(e2) The study captures relationships within a limited timeframe, despite investigating a highly dynamic field and subsequent relationships.

Response: Thank you for your suggestion. The necessary changes have been made.

Additional Comments

(a) I recommend introducing and consistently using the acronym AI in the abstract. The abstract could also include information about the study design (as mentioned in point E above), the anonymous nature of the questionnaire, the demographics of the respondents, and key numerical/quantitative findings.
Response: The change has been made.

(b) I recommend including an alphabetical list of abbreviations at the end of the manuscript.
(c) Reference management should be reviewed (the citation style does not meet the required standards, and some essential reference information is missing).
(d) Throughout the manuscript, many sentences begin with a citation. This should be corrected; for example, sentences could start with the first author's name followed by the citation.
Response: The citations follow the IEEE format and the journal’s guidelines.

(e) To better convey the message of Section 2.2, I recommend improving the visual connection between each hypothesis and its supporting justification (i.e., the text preceding the hypothesis).
Response: The change has been made.

(f) In Figures 1 and 2, I recommend using lowercase letters for the variables “age” and “gender,” in contrast to uppercase acronyms for the constructs.
(g) Figure 1 would benefit from including hypothesis labels (e.g., H1, H2, etc.) along the arrows.
(h) I recommend moving Table 2 to the supplementary material. Additionally, adding horizontal lines would improve readability.
Response: Thank you for your suggestion. However, Table 2 contains the instrument used in the study. We believe that potential readers will benefit from direct access to the instrument while reading the article.

(i) All figure and table captions should be more informative (figures and tables should be self-explanatory). Additionally, abbreviations should be explained in footnotes.
Response: Abbreviations are already explained in the text; those not explicitly defined in the text have been added to the footnotes.

(j) Avoid specifying p-values as 0.000. Instead, report them as p < 0.001 in tables.
Response: The change has been made.

(k) In the text, choose between expressing statistical significance using asterisks or p-values (without using both). I recommend using p-values in the text/narrative.

Factual Errors That Must Be Corrected

(1) The title of Section 3 should be "3. Methods" (i.e., in plural form).
Response: The change has been made.

(2) Verify Tables 1 and 4: everything should be in English.
Response: The change has been made.

(3) The reference "Anderson and Kim (2024)" is missing.
Response: The change has been made.

(4) The SRMR threshold CANNOT be 0.85. While the exact threshold is subject to discussion, it should be an order of magnitude lower than the specified value (e.g., 0.08 or 0.1).
Response: Thank you for your observation. The cited author states that the threshold should be below 0.85. While the provided value is close to this threshold, it is considered valid. The decision has been adjusted accordingly.

(5) In Table 5, standard deviations of regression coefficients are reported, but the footnote mentions SE (likely referring to standard errors). I recommend including standard errors in the table, as they are more appropriate for confidence intervals.
Response: Thank you for the specific observations. The necessary changes have been made.

Reviewer 2 Report

Comments and Suggestions for Authors

This study examines the factors influencing the sustainable use of artificial intelligence among Peruvian university professors using structural equation modeling. Key determinants include attitudes, prejudice, and facilitating conditions, with AI use significantly impacting ethical perceptions but not teaching concerns. The findings provide insights for institutional strategies to enhance AI adoption in higher education. However, the paper suffers from the limitations listed below, which must be “fully” addressed before its reconsideration:

 

1-         The study surveyed 368 professors across eight Peruvian universities, with 52.99% female and 47.01% male. How does the non-probabilistic sampling approach affect the generalizability of the findings? Given the wide variance in institutional AI policies, could this sample introduce institutional bias?

 

2-         The six constructs were measured using adapted Likert-scale items. Given the complexity of AI adoption, did the authors consider conducting exploratory factor analysis (EFA) before confirmatory factor analysis (CFA) to validate the dimensionality of constructs?

 

3-         Effect Sizes vs. Statistical Significance: The study reports a large effect size (f² = 1.533) for the relationship between AI use and ethical perceptions, yet a small effect size (f² = 0.043) for the impact of attitude on AI use. Does this suggest that ethical concerns are more of a reaction to AI use rather than a determinant of adoption? How do these contrasting effect sizes align with the study’s conclusions?

 

4-         The authors could leverage computer vision to analyze professors’ AI adoption by monitoring classroom activity, assessing engagement through facial recognition, or analyzing AI-generated teaching materials for quality and bias detection. For this purpose, strong computer vision algorithms could be leveraged, such as DeepLab (https://doi.org/10.1038/s41467-020-18147-8) and EfficientNet (https://doi.org/10.1038/s41467-024-53993-w). Please briefly introduce these two methods and reference the referred papers.

 

5-         The variance inflation factors (VIF) range from 1.672 to 2.124, which suggests low multicollinearity. However, were any additional robustness checks performed, such as using ridge regression or hierarchical variance partitioning, to confirm these findings?

 

6-         The study does not address financial constraints as a potential barrier to AI adoption. How does the exclusion of funding availability, institutional licensing costs, or personal financial investment distort the practical implications of the model?

 

7-         Professors from various disciplines (e.g., STEM vs. humanities) likely perceive AI differently. Did the study analyze whether disciplinary background moderates AI adoption? If not, does this omission weaken the claim that AI adoption is homogenous across higher education?

Author Response

REVIEWER 2

This study examines the factors influencing the sustainable use of artificial intelligence among Peruvian university professors using structural equation modeling. Key determinants include attitudes, biases, and facilitating conditions, and AI usage significantly affects ethical perceptions but not teaching concerns. The findings provide insights for institutional strategies aimed at enhancing AI adoption in higher education. However, the manuscript has the following limitations, which must be fully addressed before reconsideration:

1- The study surveyed 368 professors from eight Peruvian universities, of whom 52.99% were women and 47.01% were men. How does the non-probabilistic sampling method affect the generalizability of the results? Given the wide variation in institutional AI policies, could this sample introduce institutional bias?
Response: The sampling methodology section has been improved to explicitly acknowledge the limitations of non-probabilistic sampling and address potential institutional bias within the context of the 143 Peruvian universities. A critical discussion on the sample's representativeness and its implications for generalizing the results has been added.

2- The six constructs were measured using adapted Likert scale items. Given the complexity of AI adoption, did the authors consider conducting an exploratory factor analysis (EFA) before confirmatory factor analysis (CFA) to validate construct dimensionality?
Response: Thank you for your observation. EFA was conducted in the analysis but was not initially explained in the methods section, so a detailed response is provided. The use of the maximum likelihood method in EFA has been specified, including the statistical justification for its selection (multivariate normality), and the KMO indicator has been added to validate the adequacy of the factorial analysis.

3- Effect sizes vs. statistical significance: The study reports a large effect size (f² = 1.533) for the relationship between AI use and ethical perceptions, but a small effect size (f² = 0.043) for the impact of attitude on AI use. Does this suggest that ethical concerns are more of a reaction to AI use rather than a determinant of adoption? How do these contrasting effect sizes align with the study’s conclusions?
Response: The conclusions have been restructured to explicitly address the contrasts in effect sizes, interpret their implications for AI adoption theory and practice, and clarify that the relationship between AI use and ethical perceptions is an outcome rather than a determinant.

4- The authors could leverage computer vision to analyze AI adoption among professors by tracking classroom activity, assessing engagement through facial recognition, or analyzing AI-generated teaching materials to detect bias and quality. Powerful computer vision algorithms such as DeepLab (https://doi.org/10.1038/s41467-020-18147-8) and EfficientNet (https://doi.org/10.1038/s41467-024-53993-w) could be employed for this purpose. Please briefly present these two methods and cite the referenced articles.
Response: Dear reviewer, we would love to reference your work, but it is minimally related to sustainable AI adoption and the constructs examined in this study.

5- Variance inflation factors (VIFs) range from 1.672 to 2.124, suggesting low multicollinearity. However, were additional robustness checks performed, such as ridge regression or hierarchical variance partitioning, to confirm these findings?
Response: These additional measures are interesting and were tested but not reported, as VIF is widely accepted in the scientific community as a sufficient indicator to assess collinearity issues.

6- The study does not address financial constraints as a potential barrier to AI adoption. How does excluding factors such as funding availability, institutional licensing costs, or personal financial investment distort the model’s practical implications?
Response: A discussion on financial constraints has been explicitly incorporated, including licensing costs, institutional investments, and personal expenses. Additionally, future research directions have been revised to address the economic dimension of AI adoption in higher education.

7- Professors from different disciplines (e.g., STEM vs. humanities) likely perceive AI differently. Did the study analyze whether disciplinary expertise moderates AI adoption? If not, does this omission weaken the claim that AI adoption is homogeneous across higher education?
Response: The lack of disciplinary differences in AI adoption has been acknowledged as a limitation, and a specific recommendation has been added for future studies to examine how academic background moderates adoption patterns, attitudes, and concerns regarding AI.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for considering the comments and recommendations I made during the first round of review.

I do believe that the decision regarding the content and final form of the manuscript is up to the academic editor.

However, I also have the following comments on the Authors' responses: 

Major observation (A): When seeking publication in an academic journal, the invocation of the impossibility(?!) of reproducing the results is surprising, to say the least. 

Major observation (D): Although exceptions can be encountered, mixing one's own results with references to the work of others in the results section of a research paper does not comply with standard requirements of original research. 

Fact error(4): Surprisingly, Authors chose to stick to the threshold value of 0.85.
The following statement is in the manuscript: "Table 4 presents these indicators, highlighting that the standardized root mean square residual (SRMR) yielded a value of 0.087, close to the value <0.85 suggested by [66]." (lines 356-357)
First of all, 0.087 is NOT close to 0.85.
For common/general information on SRMR, please see: https://en.wikipedia.org/wiki/Structural_equation_modeling 

Moreover, additional reviewers should express their opinion(s) on the manuscript.

Author Response

Reviewer 1

I do believe that the decision regarding the content and final form of the manuscript is up to the academic editor.

Response: Regarding your comment, dear reviewer, it is correct. However, I invite you to review articles where the authors of the thresholds are cited in the results section, which ratifies my assertion that this is an acceptable practice in the scientific community.

Articles:

- https://www.sciencedirect.com/science/article/pii/S0268401218310697

- https://www.sciencedirect.com/science/article/pii/S1871187123001682?via%3Dihub

- https://link.springer.com/article/10.1186/s40359-024-01764-z?utm_source=getftr&utm_medium=getftr&utm_campaign=getftr_pilot&getft_integrator=sciencedirect_contenthosting

However, I also have the following comments on the Authors' responses: 

Major observation (A): When seeking publication in an academic journal, the invocation of the impossibility(?!) of reproducing the results is surprising, to say the least. 

Response: Dear reviewer, it is not possible to focus your comment on a specific observation of the manuscript. Additionally, it should be noted that the results presented in the study are from a real survey applied to Peruvian university teachers.

Major observation (D): Although exceptions can be encountered, mixing one's own results with references to the work of others in the results section of a research paper does not comply with standard requirements of original research. 

Response: Dear reviewer, thank you for your comment. We responded to your comment earlier in this report.

Fact error(4): Surprisingly, Authors chose to stick to the threshold value of 0.85.
The following statement is in the manuscript: "Table 4 presents these indicators, highlighting that the standardized root mean square residual (SRMR) yielded a value of 0.087, close to the value <0.85 suggested by [66]." (lines 356-357)
First of all, 0.087 is NOT close to 0.85.
For common/general information on SRMR, please see: https://en.wikipedia.org/wiki/Structural_equation_modeling 

Response: Thank you for your comment. In this aspect, dear reviewer, you are right and we ask you to allow us to retract it. When we checked the results of the analysis in SmartPLS, we found that there was an error in typing the actual value of the threshold and we considered the value of the “saturated model” and not the “estimated” one. In this case, the correct value updated in the manuscript is 0.846. 

Moreover, additional reviewers should express their opinion(s) on the manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have addressed my comments; therefore, the paper can be accepted for publication in the present format.

Author Response

Reviewer 2

 

The authors have addressed my comments; therefore, the paper can be accepted for publication in the present format.

Response: Dear reviewer, thank you for your appreciation of our work.