Generative AI and Cognitive Challenges in Research: Balancing Cognitive Load, Fatigue, and Human Resilience

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I have read the manuscript submitted by the authors and overall believe that the research topic demonstrates novelty and practical significance. The study investigates the application of generative artificial intelligence (GenAI) in academic research, focusing on its relationship with cognitive load, task fatigue, research engagement, resilience, and research quality. Based on a relatively large sample of 998 scholars, the authors employed SEM-PLS modeling along with mediation and moderation analyses to present the data relationships. The methodological framework is reasonably complete, and the findings provide meaningful implications for academic institutions, policymakers, and technology developers. Particularly noteworthy is the finding that high levels of GenAI immersion do not necessarily reduce cognitive load; instead, they may exacerbate declines in research quality. This result challenges the common assumption that AI inherently eases workload, thus offering valuable insights for further discussion.

However, the manuscript still requires moderate revisions to strengthen its persuasiveness and readability. First, while the authors highlight the study’s contributions, the distinctions from existing literature and the added value remain insufficiently clear. It is recommended that the introduction and discussion sections more explicitly compare prior studies and emphasize the theoretical and practical significance of the present work. In this context, I would suggest that the authors introduce a recent paper entitled “Evolutionary Game Analysis of Artificial Intelligence Such as the Generative Pre-Trained Transformer in Future Education” (https://doi.org/10.3390/su15129355) in the Introduction section, as it highlights the application prospects of generative AI in the education domain and may enrich the contextualization of this study.

Second, the data collection relies heavily on online surveys and self-reported measures. Although the authors briefly acknowledge potential bias, the limitations are discussed only superficially. It is advisable to expand this section and explain in more detail how issues such as social desirability bias and sample representativeness were addressed or controlled. Third, while the research framework and statistical modeling are comprehensive, parts of the results interpretation are overly lengthy, with excessive reporting of statistical indicators but insufficient synthesis. It would be helpful to maintain the technical rigor while providing more concise summaries to facilitate comprehension by readers without advanced statistical backgrounds. Fourth, although the discussion addresses implications for policy and practice, the argument remains largely theoretical. The authors are encouraged to add more concrete examples, such as how universities could design AI usage guidelines or training programs, to strengthen the practical relevance of the findings.

In terms of language and structure, the manuscript is relatively lengthy, and certain sections repeat similar ideas, which undermines the overall coherence. A more concise presentation, particularly in the literature review and results interpretation, would improve readability and logical flow. Furthermore, while the references are broad, some recent international studies have not been cited, and their inclusion could enhance the academic currency of the work.

Author Response

Enclosed File

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This study looks at how cognitive load and fatigue affect researchers’ work quality and influence, with engagement and resilience helping as buffers, and GenAI immersion acting like a “volume knob” on those effects. With PLS-SEM on 998 researchers, the surprising bit is: heavier GenAI use can actually make the bad effects of load/fatigue worse (useful to know, and very practical). The interaction plots are clear, and reliability looks strong; but the sampling and some measurement choices mean we shouldn’t read it as causal truth. Biggest impact here is guidance for policy/training on more responsible, balanced GenAI use in research.

Comments on the Quality of English Language

Typos & duplicates

• “advace” should be “advance”.
• “is is aligns” should be “it aligns” (or “this aligns”).
• Remove duplicated words elsewhere (scan for “the the”, “in in”).

Articles & determiners

• “The cognitive load is…” → “Cognitive load is…”.
• “At the high fatigue” → “At high fatigue levels”.

Sentence length: Break long sentences with multiple clauses into two. Example:

“Given high cognitive load and task fatigue, GenAI immersion strengthens negative outcomes which suggests…”

Author Response

Enclosed File

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

By articulating the phenomenon of cognitive workload and stress with the use of GenAI tools in research, the present paper sheds the light on fundamental issues, such as (i) the contribution of these tools to boost productivity while maintaining high performance levels (ii) their role alleviating work flow and the user's stress (iii) by freeing the user from repetitive and burocratic tasks, fostering their creativity and innovation capacity.

The paper should extend its conclusions to other domains that not only research, as its conclusions probably apply to most fields.

Highlighting the importance of the user's education is fundamental. 

Author Response

Enclosed File

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

Este artículo de investigación examina la interacción entre la carga cognitiva, la fatiga por tareas repetitivas, la resiliencia individual y la inmersión en herramientas de IA generativa (GenAI) (como ChatGPT o Elicit), analizando cómo estos factores afectan la calidad e influencia de la investigación académica. Los autores utilizaron una muestra de 998 investigadores de prestigiosas universidades de todo el mundo y aplicaron el Modelado de Ecuaciones Estructurales con Mínimos Cuadrados Parciales (SEM-PLS). El estudio analiza cómo todos estos factores están interrelacionados y cómo se moderan entre sí en los entornos académicos contemporáneos. La investigación se basa en la Teoría de la Carga Cognitiva (CLT), que explica cómo la sobrecarga mental reduce el aprendizaje y el rendimiento. Los autores recopilaron datos a través de cuestionarios en línea, utilizando un muestreo estratificado según el nivel de uso de IA, y evaluaron la validez y la confiabilidad a través de parámetros como AVE, HTMT y bootstrapping. Algunos de los principales hallazgos indican que la dependencia excesiva de GenAI reduce la originalidad y el pensamiento crítico, lo que a su vez disminuye la calidad de la investigación, especialmente en condiciones de sobrecarga cognitiva. El estudio también destaca que la tecnología no puede reemplazar la motivación y la resiliencia humanas; Su uso debe ser deliberado, equilibrado y crítico. Esta investigación subraya que la IA generativa puede mejorar la productividad con un uso moderado, pero un uso intensivo sin estrategias cognitivas adecuadas puede tener efectos contraproducentes. Finalmente, los autores proponen desarrollar sistemas adaptativos que detecten la carga cognitiva y ajusten el apoyo en consecuencia, junto con políticas institucionales sobre alfabetización en IA y directrices éticas claras. También recomiendan estudios longitudinales para identificar el punto de inflexión en el que la IA pasa de ser beneficiosa a ser perjudicial.

El artículo presenta contenido interesante, relevante y novedoso, en particular gracias a su enfoque moderador de la IA generativa en variables cognitivas. Gracias a las mejoras metodológicas y formales, y al fortalecer su marco teórico y su discusión, tiene un gran potencial para su publicación en una revista indexada en el JCR en los campos de las ciencias sociales aplicadas, la educación o la tecnología.

However, several areas need improvement to increase its rigor. In terms of formatting, there are sections or paragraphs with minor errors or inconsistencies in titles, numbering, and/or tables (these simply need careful revision). The authors could also improve the cohesion between different sections of the paper (for example, the discussion repeats results without strong conceptual integration).

The critical discussion of previous literature should be expanded. Currently, it focuses mainly on CLT and some recent studies, but the authors should incorporate references on ethical analysis of AI applications, distributed cognition, and motivational theories. Although the use of SEM-PLS is appropriate, the authors should better justify their choice over other methods such as CB-SEM, and more clearly detail the sampling criteria and bias control procedures.

Another aspect for the authors to consider is applying complementary analyses, such as multigroup or invariance techniques, to strengthen external validity. They should also aim to integrate the results more explicitly with cognitive and technology adoption theories, rather than simply describing empirical findings, and clearly highlight the theoretical and practical contributions compared to the existing literature—ideally through a dedicated table or section.

Author Response

Enclosed File

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The paper studies how cognitive load and task fatigue relate to research quality/influence, with engagement and resilience as mediators and GenAI immersion as a higher-order moderator. Using PLS-SEM on 998 researchers, they find negative direct paths from load/fatigue and—more striking—GenAI immersion intensifies those negatives rather than buffering them; engagement/resilience help but only partly. Methods/validity reporting and figures are much clearer in v2, and the moderation plots make the main message easy to see.

Strengths

• Transparent methods: justified PLS-SEM, 5,000-sample bootstrapping, full reliability/validity (α/CR/AVE/HTMT) and fit (R²/Q²/SRMR).
• Clear results: hypothesis table with effect sizes, plus readable interaction plots.
• Practical angle: explicit implications, limits, and gap/implication tables for policy/training and “responsible AI use.”

Weaknesses

• Design limits: cross-sectional, self-report; keep causal wording cautious (they note this, but final claims should stay measured).
• Sampling bias risk: heavy representation from top-500 universities and social channels; acknowledged, but remind the reader in Conclusions.
• Repro details: include all item texts/scales in an appendix (Annexure III starts this; make it complete) and note any adaptations.