Triple-Entry Accounting and Other Secure Methods to Preserve User Privacy and Mitigate Financial Risks in AI-Empowered Lifelong Education

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript “Triple-Entry Accounting and other secure methods to Preserve User Privacy and Mitigate Financial Risks in AI-empowered Lifelong Education” strives to explore data privacy solutions in AI-empowered lifelong education, focusing on Triple-Entry Accounting (TEA), Merkle trees, and offline AI models.

As a general note, the topic appears relevant, but the research questions are not clearly articulated and in several sections of the paper the authors seem to divagate their discussions to general aspects of AI uses, diluting the general focus of the paper. The manuscript employs technical solutions that sound appealing but lack conceptual rigor and consistency. Discussions are sometimes overly descriptive, lack depth in critical analysis. These issues should be addressed, together with the following recommendations.

1. The Introductions should provide stronger rationale for choosing TEA and offline AI models as proposed optimal privacy solutions. The authors should also clarify the impetus for such study relying on previous case studies and better highlight the literature gaps.
2. While the manuscript references relevant studies, it lacks a systematic comparison of TEA against other privacy solutions, highlighting their pros and cons. The authors should critically assess the limitations of existing methods to justify the need for the proposed solutions.
3. The methodology describes TEA, Merkle trees, and offline models but lacks details on evaluation criteria, data sources, and experimental design. The methodology relies on a conceptual inconsistency: Triple-Entry Accounting (TEA) is inherently designed for online use, particularly because it relies on Distributed Ledger Technology (DLT), such as blockchain, for maintaining transparent and immutable records. The integrity and security of TEA come from its decentralized verification process, which requires network connectivity for cryptographic validation and consensus mechanisms. The article proposes an offline AI-tutor model alongside TEA, but TEA’s core functionality depends on an online, distributed network for recording and verifying transactions, thus creating a conceptual inconsistency since offline models inherently lack connectivity to the very network required for TEA to function.
4. The manuscript seems to suggest that TEA can preserve user anonymity, which contradicts TEA’s requirements for transparent and identifiable transactions to ensure authenticity and security. The authors should clearly acknowledge that TEA ensures authenticity by verifying all parties involved, not by anonymizing them.
5. The offline solution appears to contradict the premise of using TEA, which requires online, distributed verification. Additionally, the argument for offline use lacks depth and underestimates the challenges of updating educational content.
6. Figure 1 appears pixelated and should be recalibrated for clarity. The black background in Figures 4 and 5 should be removed.
7. Results are descriptive and lack quantitative analysis or benchmarks against other privacy methods, leading to weak evidence supporting the proposed solutions. Include quantitative results and performance benchmarks against alternative privacy-preserving methods.
8. The discussion touches on ethical implications but fails to connect the findings to the research question and practical implementation challenges. The authors should consider Including a section on real-world implementation challenges, such as scalability, cost, and usability.
9. The conclusions overstate the efficacy of offline models and TEA without sufficiently acknowledging their limitations and contradictions.

This manuscript presents an interesting mix of privacy-preserving solutions but suffers from conceptual inconsistencies, particularly with TEA’s transparency and the offline model’s incompatibility with distributed verification. Strengthening the theoretical foundation, correcting contradictions, and enhancing methodological rigor would significantly improve the study's credibility and impact.

Author Response

We would like to sincerely thank, and express our gratitude to the reviewer for the time devoted in doing a meticulous review of our work and for pointing out the parts of our manuscript that need to be improved. We carefully followed the suggested directives, and we submit our manuscript changes as follows:

Comment 1:

The Introductions should provide stronger rationale for choosing TEA and offline AI models as proposed optimal privacy solutions. The authors should also clarify the impetus for such study relying on previous case studies and better highlight the literature gaps.

Answer 1:
We significantly altered the Introduction to improve the rationale as proposed (paragraphs 2,3,4). We also clarified the impetus for authoring our study and highlighted the literature gaps (a new paragraph by the end of the Introduction section). 

Comment 2:

While the manuscript references relevant studies, it lacks a systematic comparison of TEA against other privacy solutions, highlighting their pros and cons. The authors should critically assess the limitations of existing methods to justify the need for the proposed solutions.

Answer 2:
We followed the advice of the reviewer, and we presented a systematic comparison of Triple Entry Accounting (TEA) against existing solutions in the "Materials & Methods" section:

Comment 3:

The methodology describes TEA, Merkle trees, and offline models but lacks details on evaluation criteria, data sources, and experimental design. The methodology relies on a conceptual inconsistency: Triple-Entry Accounting (TEA) is inherently designed for online use, particularly because it relies on Distributed Ledger Technology (DLT), such as blockchain, for maintaining transparent and immutable records. The integrity and security of TEA come from its decentralized verification process, which requires network connectivity for cryptographic validation and consensus mechanisms. The article proposes an offline AI-tutor model alongside TEA, but TEA’s core functionality depends on an online, distributed network for recording and verifying transactions, thus creating a conceptual inconsistency since offline models inherently lack connectivity to the very network required for TEA to function.

Answer 3:
We agree that there is a lack of evaluation criteria and data sources. This is since our approach is (currently) mainly theoretical, and no actual real-world data were collected to ensure that everything works as planned, so this study should be addressed purely as Proof of Concept. However, we provided the experimental setup details where needed (ie: TEA & DLT, Offline models). We also ensured that the methodologies work, and we verified the reproducibility of the methods we presented. We also addressed the problem of conceptual inconsistency by removing the incorporation of offline model and TEA integration from the manuscript, since it was indeed confusing. The three methods are now presented as distinct concepts. Finally, we added a comparative analysis of all three methods in the end of “Results” Section, to address and separate the strong and weak points of each method. It is our hope that this research draws the necessary attention in the future, and that we will be able to implement (and thoroughly test) all three methods in real-world conditions (provided we have the resources for such a venture).                                               

Comment 4:

The manuscript seems to suggest that TEA can preserve user anonymity, which contradicts TEA’s requirements for transparent and identifiable transactions to ensure authenticity and security. The authors should clearly acknowledge that TEA ensures authenticity by verifying all parties involved, not by anonymizing them.

Answer 4:
We agree that this issue was not properly and clearly presented in the study. We reviewed thoroughly the manuscript and altered all the relative concepts where needed so that it is clearly stated that TEA does not provide anonymity but pseudonymity.

Comment 5:

The offline solution appears to contradict the premise of using TEA, which requires online, distributed verification. Additionally, the argument for offline use lacks depth and underestimates the challenges of updating educational content.

Answer 5:
We addressed the contradiction by separating the TEA method from the offline LLM tutor (please see our answer in Comment 3).

Comment 6:

Figure 1 appears pixelated and should be recalibrated for clarity. The black background in Figures 4 and 5 should be removed.

Answer 6:
We corrected all three figures.

Comment 7:

Results are descriptive and lack quantitative analysis or benchmarks against other privacy methods, leading to weak evidence supporting the proposed solutions. Include quantitative results and performance benchmarks against alternative privacy-preserving methods.

Answer 7:
We added a new table containing a quantitative analysis between proposed methods and reviewed and enhanced the Results section.

Comment 8:

The discussion touches on ethical implications but fails to connect the findings to the research question and practical implementation challenges. The authors should consider Including a section on real-world implementation challenges, such as scalability, cost, and usability.

Answer 8:
We added a table with real-world challenges that occur for each method, based on scalability, difficulty of implementation, costs and usability. This table outlines the challenges associated with each method, making it easier to compare and understand the key issues that may arise during implementation.

Comment 9:

The conclusions overstate the efficacy of offline models and TEA without sufficiently acknowledging their limitations and contradictions.

Answer 9:
We added a paragraph in the Conclusions section which addresses the challenges and contradictions.
 

 

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Editor,

I have completed my review of the manuscript and recommend rejection with major revisions before reconsideration. While the study explores an important topic on AI privacy in lifelong learning, it lacks methodological rigor, empirical validation, and comparative analysis.

Key concerns include:

• Unclear Research Problem & Justification – The study does not clearly define its objectives or rationale for selecting TEA-DLT and Merkle Trees over other privacy-preserving methods.
• Weak Literature Review – Lacks comparative analysis with existing AI privacy models and does not clearly articulate the research gap.
• Incomplete Methodology & Empirical Validation – No technical justification, encryption details, or performance metrics. Lacks experimental results or cryptographic verification.
• Weak Argumentation & Practical Relevance – No discussion on real-world feasibility, policy implications, or implementation challenges.
• Unstructured Conclusion – Fails to summarize contributions, limitations, or future research directions.

I recommend significant revisions, including empirical validation, comparative analysis, and a stronger theoretical framework, before resubmission.

Best regards,

Comments for author File: Comments.pdf

Comments on the Quality of English Language

English need to be improved 

Author Response

We would like to express our gratitude to the reviewer for the time devoted to providing a meticulous review of our work and for pointing out the parts of our manuscript that needed to be improved. We followed all the suggested directives, and we made significant revisions to our work including adding three new tables (comparative analysis, quantitative analysis and empirical analysis) on the real-world challenges each method may have. We submit our manuscript changes addressing each comment separately as follows:

Comment 1:

Unclear Research Problem & Justification – The study does not clearly define its objectives or rationale for selecting TEA-DLT and Merkle Trees over other privacy-preserving methods.

Answer 1:

We have extensively reformed and extended the introduction to improve the rationale and clear the research problem & Justification as proposed

Comment 2:

Weak Literature Review – Lacks comparative analysis with existing AI privacy models and does not clearly articulate the research gap.

Answer 2:

We acknowledge this suggestion is true. On the other hand, the field of study is relatively new. There is not a plethora of related articles regarding personal data handling and student privacy. The specific methods proposed aim to present a series of totally new approaches to mitigating the problem. However, we thoroughly enhanced the manuscript in the introduction, materials & methods and added a comparative analysis to improve the clarity of the manuscript as proposed.

Comment 3:

Incomplete Methodology & Empirical Validation – No technical justification, encryption details, or performance metrics. Lacks experimental results or cryptographic verification.

Answer 3:

We have significantly revised and enhanced our manuscript so that it conforms to the above suggestions. We present a systematic comparison of Triple Entry Accounting (TEA) against existing solutions in the "Materials & Methods" section. We also added a Quantitative Analysis of all three proposed methods. Finally, we agree on lacking Empirical Validation. Since our approach is (currently) mainly theoretical, no actual real-world data were collected to ensure that everything works as planned. However, we provided the experimental setup details where needed (ie: TEA & DLT, Offline models). We also ensured that the methodologies work, and we verified reproducibility of the methods. Finally, we addressed the real-life challenges each method may have in Table 3. It is our hope that this research draws the necessary attention, and, in the future, we will be able to test all three methods in real-world conditions (provided we have the resources for such a venture).

Comment 4:

Weak Argumentation & Practical Relevance – No discussion on real-world feasibility, policy implications, or implementation challenges.

Answer 4:

We have significantly enhanced the manuscript addressing all the issues mentioned. We have also added a table addressing the implementation challenges of each method, as proposed. We present the real-world challenges that occur for each method, based on scalability, difficulty of implementation, costs and usability. Table 3 outlines the challenges associated with each method, making it easier to compare and understand the key issues that may arise during implementation.

Comment 5:

Unstructured Conclusion – Fails to summarize contributions, limitations, or future research directions.

Answer 5:

We altered the conclusion section so that it follows the suggestion of the reviewer

 

Reviewer 3 Report

Comments and Suggestions for Authors

I really appreciate the novelty of this work.The paper discusses innovative privacy-preserving methods for AI-driven lifelong education, including Triple-Entry Accounting (TEA), Distributed Ledger Technology (DLT), Merkle trees, and Offline AI Tutors. The structure of the work is fine, and the solutions offered are all the more relevant against the backdrop of privacy concerns. Please avoid repetition if possible. U can reexplain technical concept in simple language that would attract more readers to the work. 

Author Response

We would like to thank the reviewer for the kind words and the valuable comments which we addressed thoroughly through the revised manuscript. This gives us hope and courage to continue our venture on this research field.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors addressed my recommendations by responding to my comments.

However, I recommend that the authors explicitly state in both the introduction and conclusions sections that their study lacks empirical evaluation and real-world data sources. They should clarify that their approach is primarily theoretical at this stage, with no actual real-world data collected to validate its effectiveness. As such, the study should be positioned purely as a Proof of Concept.

Author Response

We would like to thank the reviewer once more for their time devoted to examining our work for a second time. We have implemented the suggestion as prompted both in the introduction and the conclusion sections.

Reviewer 2 Report

Comments and Suggestions for Authors

Subject:Reviewer Report – Manuscript Evaluation

Dear Writers

I should like to thank you for taking trouble to revamp and resubmitting your manuscript "Triple-Entry Accounting and Other Secure Approaches to Maintaining Users' Privacy and Mitigating Financial Risks in Artificial Intelligence-Directed Lifelong Learning." I should also thank you for taking pains to respond to my previous critique in detailed ways and making extensive revisions. But upon closer scrutiny, I regret to point out that the manuscript is not yet addressing the most essential problems raised in the original critique to an extent that justifies not advising against acceptance in this form.

Key Reasons for Rejection

While the refined version has been improved in restructuring the introduction section, expanding in the literature section, and including additional tables, there remain certain core weaknesses. These weaknesses have a vast impact over the overall value and viability of the study and prevent it from meeting standards required to be published.

1. Deficiency in Empirical Verification or Performance Measures

You have both comparative and qualitative critique of methods advanced, but the paper lacks empirical validation through testing in practice, through simulations, or through case studies. Experimental setups have been outlined but not included in the context of concrete empirical outcomes to establish effectiveness and workability in practice of methods outlined.

Necessary to Consider: Experimental testing through computer simulations, comparative case studies in practice, or computational benchmarking to validate claims made.

2. Lack of Proper Comparative Study with Other Privacy-Sustaining AI Approaches

The resubmitted paper has expanded its presentation in respect to privacy-preserving AI models but does not yet have an organized comparative analysis between methods advanced in this work and traditional methods such as federated learning, homomorphic encryption, zero-knowledge proofs, or differential privacy. How TEA-DLT and Merkle Trees should be preferred to these methods is not yet developed to an appropriate extent.

Necessary to Consider: Systematic comparison to existing privacy-preserving methods in AI, focusing on strengths, weaknesses, and performance sacrifices.

3. Deficiency in Research Framework and Evaluation Measures

Despite including a table summarizing real-world problems to each technique, the paper is lacking in systematic evaluation criteria to determine privacy protection, security strength, computational efficiency, or scalability. It is not straightforward to decide upon effectiveness and viability of methods in hand in the absence of clear evaluation criteria.

The following is essential to consider: Clearly established evaluation criteria regarding security, privacy, computational cost, and scalability.

4. Narrow Practical Discourse Concerning Implementation Problems & Impacts to Stakeholders

While this manuscript has improved in addressing problems in implementation through Table 3, feasibility, institutional factors, regulatory factors, and economic factors have not been detailed to an appropriate extent. Policy contexts, partners in the industry, or implications beyond these methods in practice have not been consulted appropriately by this work.

Necessary Considerations: Dedicated section discussing institutional, regulatory, and economic viability aspects in practical application

5. Directions for Future Research and Conclusion Are Underdeveloped

The conclusion has been reordered to better summarize key outcomes, yet is lacking in specifics about future directions in research. While the authors comment that testing in reality is required, the manuscript is not explicit about next steps in exactly how this testing can be conducted.

Necessary to Consider: Properly organized conclusion with explicitly stated contributions, limitations, and elaborated future work directions

The Verdict: Rejection

Though these modifications have been done, problems raised in the original submission have not yet been resolved, most notably in empirical validation, comparative evaluation, and explicitly developed evaluation criteria. Since this is submission number two and major methodology changes have to be done yet, I have to recommend rejecting it in this manner.

In an effort to resubmit an upgraded version of this paper in the future

Conducting empirical tests, simulations, or field/real-world case studies

Presenting comparative performance benchmarks to existing privacy-preserving techniques

Setting clear evaluation standards to measure security, privacy, and effectiveness.

Taking this debate to real-world applicability, scalability, and sector significance

I appreciate that you have thought about publishing your work and encourage you to seriously take these recommendations to heart in future submissions and work. I welcome the possibility to consider an improved submission of your work that takes these necessary changes into consideration. Best wishes

Comments on the Quality of English Language

Dear Editor,

I have carefully reviewed the revised manuscript and the authors' responses. While some improvements have been made, the core concerns regarding empirical validation, comparative analysis, and evaluation metrics remain unresolved. Given that this is the second submission and these fundamental issues persist, I recommend rejecting the manuscript in its current form. The authors would need to conduct empirical testing, provide structured comparisons, and define evaluation criteria before reconsideration for publication.

Best regards,

Author Response

We hereby want to express once more our gratitude to the reviewer for the time and effort devoted to examining our work for the second time. We sincerely appreciate the reviewer's dedication to providing a thorough evaluation of our updated work and pointing out the areas requiring improvement. We have diligently implemented all suggested revisions, and our manuscript changes are outlined below:

Comment 1:

We fully agree with the comment that the paper lacks empirical validation. Since the first two suggestions (testing and simulation) will be difficult to materialize within the strict time frame that we need to respond, we have added a new section called “Empirical Validation through comparison with existing case studies” where we compare which hopefully addresses the concerns of the reviewer. The new chapter can be found at the end of “Results” section.

Comment 2:

This is indeed a very insightful suggestion for future work. Since both methods (TEA-DLT and Merkle Proofs) use Blockchain technology, we are averted from diving into such a specialized field as homomorphic encryption and ZKPs. Both methods are highly capable of solving the privacy problem, but there are no current implementations on education and the highly technical nature of both would present a big technological barrier. The combination of federated learning with homomorphic encryption does exist in research implementations, more specifically in healthcare and finance where privacy is vital. However, practical deployment in education -again- faces significant challenges due to the computational overhead of homomorphic encryption. However, we followed the suggestion of the reviewer, and we added a systematic comparison on a new table (Table 3) that addresses the strengths, weaknesses, and performance sacrifices of each method.

Comment 3:

We have clearly established evaluation criteria regarding security, privacy, computational cost, and scalability and reformed Table 3 (now Table 4) as per the reviewer’s suggestion.

Comment 4:

We have added a new section called “Practical Considerations for Privacy Implementation” as proposed by the reviewer.

Comment 5:

We added a separate section in the “conclusions” that addresses the reviewer’s suggestion.