Preamble

Research and Solutions Regarding Research Integrity Issues of Artificial Intelligence in Academic Writing

1. Introduction

In recent years, the rapid development of Artificial Intelligence (AI), particularly Large Language Models (LLMs) such as ChatGPT, has brought revolutionary changes to the field of scientific research. While these technologies significantly improve the efficiency of literature reviews, data analysis, and linguistic polishing, they also introduce unprecedented challenges to research integrity. Issues such as "AI-generated content" (AIGC) masquerading as original thought, data fabrication, and the blurring of authorship boundaries have become focal points of concern within the global academic community. This paper explores the specific manifestations of research integrity risks associated with AI in academic writing and proposes a multi-dimensional framework for mitigation.

2. Analysis of Research Integrity Risks

The application of AI in academic writing primarily poses risks in the following four dimensions:

2.1 Authorship and Attribution

One of the most contentious issues is whether an AI system can be credited as an author. According to the guidelines of COPE (Committee on Publication Ethics) and many high-impact journals, AI does not meet the criteria for authorship because it cannot take legal or ethical responsibility for the content of the work. However, the "hidden" use of AI to generate entire sections of text without disclosure constitutes a form of academic dishonesty akin to ghostwriting.

2.2 Content Authenticity and "Hallucinations"

LLMs are probabilistic models that predict the next token in a sequence; they do not possess a true understanding of factual reality. This often leads to "AI hallucinations," where the system generates plausible-sounding but entirely fabricated citations, data points, or experimental results. If researchers fail to rigorously verify AI-generated content, they risk disseminating false information, thereby undermining the credibility of the scientific record.

2.3 Data Privacy and Intellectual Property

The process of training AI models often involves vast datasets that may include copyrighted materials or sensitive data. When researchers input unpublished experimental data or proprietary ideas into public AI tools for optimization, they risk data leakage. Furthermore, the copyright status of AI-generated text remains legally ambiguous in many jurisdictions, creating potential intellectual property disputes.

2.4 Algorithmic Bias and Academic Diversity

AI models are trained on existing literature, which may contain inherent biases regarding gender, race, or geographical representation. Over-reliance on AI for literature synthesis may inadvertently reinforce these biases and lead to a homogenization of academic discourse, potentially stifling

1 王

Department of Pathology, School of Basic Medical Sciences, Jilin University

Abstract

In recent years, the rapid development of machine learning and deep learning has significantly transformed the field of biomedical research. This paper explores the integration of advanced computational techniques within the domain of pathology, focusing on their application in diagnostic accuracy and predictive modeling. By leveraging large-scale datasets and sophisticated algorithms, researchers can now identify complex patterns in histopathological images that were previously imperceptible to the human eye.

Introduction

The integration of artificial intelligence into clinical pathology represents a paradigm shift in modern medicine. Traditional diagnostic methods, while foundational, often face challenges related to inter-observer variability and the labor-intensive nature of manual slide analysis. The emergence of digital pathology has facilitated the transition from physical glass slides to high-resolution digital images, providing a fertile ground for the application of machine learning (ML) and deep learning (DL) frameworks.

Methodology

Our research methodology focuses on the implementation of convolutional neural networks (CNNs) for the automated classification of tissue samples. The process begins with the acquisition of whole-slide images (WSIs), which are subsequently pre-processed to ensure consistency in staining and resolution.

[FIGURE:1]

As illustrated in [FIGURE:1], the workflow involves several critical stages: data normalization, feature extraction, and model training. We utilize the function $\mathcal{F}(x)$ to represent the mapping between the input image data and the predicted diagnostic category. The optimization of the network parameters is achieved by minimizing the loss function $L$, defined as:

$$L = \sum_{i=1}^{n} (y_i - \hat{y}_i)^2 + \lambda |\omega|^2$$

where $y_i$ denotes the ground truth label, $\hat{y}_i$ represents the predicted value, and $\lambda |\omega|^2$ serves as the regularization term to prevent overfitting.

Results and Discussion

The experimental results demonstrate that the proposed deep learning model achieves high sensitivity and specificity across various pathological subtypes. As shown in [TABLE:1], the accuracy of the automated system is comparable to, and in some instances exceeds, that of senior pathologists.

[TABLE:1]

Furthermore, the application of these models extends beyond simple classification. By analyzing the spatial distribution of tumor-infiltrating lymphocytes (TILs), we can derive prognostic indicators that are crucial for personalized treatment planning. These findings are consistent with previous studies \cite{1

摘要

Research Integrity Risks Induced by AI-Assisted Academic Writing and Proposed Countermeasures

Introduction

In recent years, the rapid development of artificial intelligence (AI), particularly Large Language Models (LLMs), has revolutionized the landscape of scholarly communication. While AI-assisted academic writing offers significant benefits in terms of efficiency, language polishing, and data synthesis, it simultaneously introduces profound challenges to research integrity. The boundary between legitimate assistance and academic misconduct has become increasingly blurred, necessitating a rigorous examination of the associated risks and the development of robust countermeasures.

1. Research Integrity Risks in AI-Assisted Writing

The integration of AI into the research workflow presents several critical risks that threaten the foundational principles of scientific honesty and accountability.

1.1 Hallucination and Data Fabrication

One of the most significant technical limitations of current LLMs is the phenomenon of "hallucination." AI systems may generate plausible-sounding but entirely fictitious citations, data points, or experimental results. When researchers rely on these outputs without rigorous verification, they inadvertently introduce fabricated content into the scientific record, undermining the reliability of published research.

1.2 Plagiarism and Attribution Challenges

AI models are trained on vast datasets of existing literature. Consequently, the text they generate may closely mirror original works without providing proper attribution. This leads to "mosaic plagiarism" or "AI-generated plagiarism," where the lack of a clear trail of authorship makes it difficult for traditional similarity-detection software to identify the source. Furthermore, the question of whether AI-generated content itself constitutes original work remains a subject of intense legal and ethical debate.

1.3 Erosion of Critical Thinking and Authorship

Over-reliance on AI for drafting arguments and interpreting data can lead to a "black box" approach to scholarship. If researchers delegate the core intellectual task of synthesis and critical analysis to an algorithm, the essence of authorship—intellectual responsibility—is compromised. This raises concerns regarding who is ultimately accountable for the accuracy and validity of the findings presented in a paper.

1.4 Bias and Lack of Transparency

AI models often reflect the biases present in their training data. In academic writing, this can manifest as skewed literature reviews or the reinforcement of specific theoretical paradigms while ignoring marginalized perspectives. Moreover, the lack of transparency regarding the extent of AI involvement in a manuscript makes it difficult for peer reviewers and readers to assess the potential for algorithmic bias.

2. Proposed Countermeasures

To mitigate these risks and ensure the ethical use of AI

方法

Dishonesty in Academic Writing and Literature Reviews: Designing a Prevention Framework through Ethics, Education, and Institutional Policy

Introduction

In the contemporary academic landscape, the integrity of scholarly writing and literature reviews is fundamental to the advancement of knowledge. However, the proliferation of academic misconduct—ranging from subtle plagiarism to the sophisticated manipulation of data—poses a significant threat to the credibility of scientific research. Dishonesty in writing and reviewing literature not only undermines the individual researcher's reputation but also erodes public trust in science and wastes valuable institutional resources. To address these challenges, it is essential to move beyond reactive measures and develop a proactive, multi-dimensional prevention framework. This paper proposes a comprehensive strategy for safeguarding academic integrity by integrating three critical dimensions: ethical cultivation, pedagogical intervention (education), and institutional policy declarations.

1. Manifestations of Dishonesty in Academic Writing and Reviews

Dishonesty in academic writing often manifests in diverse and increasingly complex ways. Beyond blatant plagiarism, researchers may engage in "salami slicing" (fragmenting a single study into multiple small papers), selective reporting of results to support a specific hypothesis, or the fabrication of citations to bolster weak arguments. In the context of literature reviews, misconduct often takes the form of biased selection—deliberately omitting contradictory evidence or over-representing the work of colleagues and mentors (cronyism). Furthermore, the rise of generative AI has introduced new risks, such as the "hallucination" of non-existent sources and the use of automated tools to paraphrase stolen ideas without proper attribution. These behaviors distort the scholarly record and hinder the cumulative nature of scientific progress.

2. The Ethical Dimension: Cultivating Internalized Integrity

The first pillar of the prevention framework focuses on the ethical development of the researcher. Academic integrity should not be viewed merely as a set of rules to be followed to avoid punishment, but as a core professional value.

• Internalization of Virtue: Ethical training must emphasize the "internal morality" of science, fostering virtues such as honesty, transparency, and intellectual humility. Researchers should be encouraged to view their work as a contribution to a collective pursuit of truth rather than a mere means of career advancement.
• Reflective Practice: Promoting a culture of self-reflection allows scholars to recognize their own biases and the systemic pressures that might tempt them toward shortcuts. By prioritizing the quality and reliability of work over the quantity of publications, the ethical dimension addresses the root psychological drivers of misconduct.

3

结果

Comprehensive Governance Solutions

Current conceptual frameworks for addressing research misconduct, such as data fabrication, still lack rigorous cross-disciplinary empirical validation. Furthermore, existing literature has not sufficiently addressed the critical impact of the rapid iteration speed of generative AI on the timeliness and effectiveness of regulatory oversight.

结论

Multi-agent collaborative governance can safeguard academic integrity and public trust in the era of Artificial Intelligence.

关键词

Research Integrity Challenges and Artificial Intelligence in Academic Writing

Introduction

The rapid advancement of artificial intelligence (AI), particularly Large Language Models (LLMs), has revolutionized the landscape of academic research and dissertation writing. While these tools offer unprecedented efficiency in data processing, literature synthesis, and linguistic refinement, they simultaneously introduce significant challenges to research integrity. As AI becomes increasingly integrated into the scholarly workflow, the academic community must address emerging concerns regarding authorship, transparency, and the potential for automated misconduct.

The Impact of AI on Research Integrity

The integration of AI into academic writing presents a dual-edged sword. On one hand, machine learning algorithms can assist researchers in identifying patterns within vast datasets and improving the clarity of technical prose. On the other hand, the ease with which AI can generate human-like text poses a risk to the fundamental principles of originality and accountability.

Key issues include:

• Authorship and Attribution: Determining the extent of AI contribution remains a contentious issue. Current academic standards emphasize that AI cannot be credited as an author, yet the boundary between "AI-assisted editing" and "AI-generated content" is often blurred.
• Data Fabrication and Hallucination: LLMs are known to produce "hallucinations"—plausible-sounding but entirely factually incorrect statements or citations. In a research context, this can lead to the inadvertent inclusion of fabricated data or non-existent references, undermining the reliability of the dissertation.
• Plagiarism and Patchwork Writing: While traditional plagiarism detection tools are evolving, AI can generate "original" phrasing that bypasses standard similarity checks while still misappropriating the underlying ideas or structures of existing work.

Solutions and Mitigation Strategies

To preserve the sanctity of the doctoral dissertation and scholarly publications, institutional and technological solutions must be implemented.

1. Policy and Disclosure Frameworks

Universities and journals must establish clear guidelines regarding the use of AI. This includes mandatory disclosure statements where authors specify which AI tools were used and for what purpose (e.g., grammar correction versus content generation). Transparency is the cornerstone of maintaining trust in the peer-review process.

2. Advanced Detection and Verification

The development of AI-detection software is a critical technical response. However, detection alone is insufficient. Academic institutions should prioritize "process-based" evaluation, where supervisors engage more deeply with the student's writing stages to ensure the intellectual work remains human-led.

3. Enhancing AI Literacy

Educating researchers

1 Hou

Xiangyi

1 Wang

Yishu Lisha (Department Pathology, College Basic Medical Sciences, Jilin University, Changchun, 130000)

Abstract

Abstract

[Objective] This study aims to investigate the research integrity risks triggered by AI-assisted academic writing and propose comprehensive countermeasures to address these challenges.

[Methods] We systematically cataloged dishonest behaviors occurring during the writing and peer-review processes. Based on this analysis, we designed a prevention framework that spans ethical education, technical oversight, and institutional disclosure requirements.

[Results] A governance package integrating ethical awareness, technological tools, and regulatory rules was constructed. This framework is designed to reduce the incidence of ghost-writing, data fabrication, and other forms of academic misconduct associated with generative AI.

[Limitations] The proposed countermeasures require further cross-disciplinary empirical validation. Additionally, the impact of rapid iterations in generative AI technology on regulatory timeliness has not yet been fully addressed.

[Conclusions] Multi-stakeholder collaborative governance is essential to safeguard academic truth and maintain public trust. This paper represents one of the research outcomes dedicated to these objectives.

Keywords

Artificial intelligence Dissertation Writing Research Integrity Problems

1 引言

Academic writing is a fundamental component of scientific research, characterized by the structured expression of ideas, dynamic argumentation, and rigorous logical reasoning. With the rapid advancement of Artificial Intelligence (AI) technology, its application within the academic sphere—particularly in paper writing—has increased significantly. Numerous studies have explored the effectiveness of AI-based writing tools in higher education environments. For instance, Miranty and Almusharraf investigated the impact of the AI-based grammar checker, Grammarly, on the writing skills of undergraduate students. Their findings indicated that, compared to a control group, students using Grammarly demonstrated improvements in both grammatical accuracy and overall writing quality. Furthermore, Farrokhnia and Rospigliosi demonstrated how ChatGPT could be utilized to assist graduate students in generating research proposals, noting that the generated content was particularly valuable for providing preliminary ideas and conceptualizing research topics.

By continuously learning and optimizing through dialogue with researchers, these tools assist in conducting scientific research and drafting manuscripts. Specific functions include: (1) Brainstorming: Utilizing big data processing capabilities to achieve rapid retrieval, recommendation, and integration of existing data, thereby providing inspiration, source material, or informational support to uncover deeper research content and ideas; (2) Content Generation and Editing:

Rapidly generating text according to specific requirements to assist in the completion of abstract writing.

关键词

Polishing, machine translation, and related tasks; ( ) Literature Review: Leveraging cross-disciplinary and multimodal knowledge association capabilities to assist researchers in gaining a more comprehensive understanding of knowledge and rapidly extracting core content.

As a new driven writing tool, it offers numerous conveniences. However, large-scale user exploration has gradually revealed the technical boundaries of these algorithms. The resulting transformation of research paradigms poses new challenges to research integrity based on existing frameworks. Due to the lack of transparency in the underlying implementation and usage processes of , the generated content often lacks explicit citations of data sources. This not only compromises the transparency of scientific research but also raises further concerns regarding the authenticity and reliability of the generated content.

is merely a statistical model and is incapable of generating truly original ideas. Consequently, it cannot fully or deeply comprehend the meaning of the prompts it receives, which may lead to the generation of information that appears plausible but is actually false. For example, in a proof-of-concept study conducted by Martin, ChatGPT—powered by the GPT-3 language model—was used to generate a fraudulent article related to the field of neurosurgery. Following its generation, experts in neurosurgery, psychiatry, and statistics reviewed the article for accuracy and coherence, comparing it with existing similar papers. While the fraudulent article created by ChatGPT resembled a genuine scientific paper in terms of word choice, sentence structure, and overall composition, significant problems and errors emerged during the generation process, particularly regarding bibliographic citations. During the evaluation of ChatGPT, researchers observed that even when provided with fictional information, the language model would continue to generate associated false content following the questioner's logic. Given ChatGPT's ability to generate coherent and plausible sentences, practitioners in the field of education are increasingly concerned that this technology may adversely affect the development of independent thinking and writing skills in graduate students. Several universities have already taken measures to address this challenge: faculty at New York University have classified the use of ChatGPT as academic misconduct and explicitly prohibited it in their syllabi; similarly, the University of Hong Kong emphasized via an internal email to all faculty and students that the use of ChatGPT is strictly forbidden in all classroom tasks, assignments, and assessments. Conversely, if artificial intelligence technologies represented by ChatGPT are utilized rationally, they could significantly promote the cultivation of critical thinking and innovative capabilities in graduate students. Therefore, this paper aims to discuss the various research integrity issues that may arise during the process of thesis writing and literature review and proposes a series of

strategies to mitigate the challenges encountered when is involved in academic writing, striving toward the establishment of a healthy and robust academic environment.

2.1 人工智能写作工具概述

According to the definition provided in the 2022 Artificial Intelligence Generated Content (AIGC) White Paper, the application of AI technology in academic paper writing can be categorized into two types based on the creative methodology: imitation-based creation and concept-based creation. Simply put, imitation-based academic writing involves analyzing and studying the characteristic patterns of existing human-authored papers to develop models that follow these established rules. In contrast, the concept-based academic writing mode is not limited to the observation and emulation of specific data. Instead, it extracts abstract conceptual keywords from massive datasets and recombines different textual elements to produce innovative academic outcomes.

Currently, the most commonly used AI writing tools in this field include ChatGPT and DeepSeek. ChatGPT (Generative Pre-trained Transformer) is an AI chatbot program developed by the American research laboratory OpenAI. As a natural language processing technology based on deep learning, ChatGPT's popularity continues to rise, particularly with its latest iterations, GPT-4 and GPT-4o. These models possess robust text generation capabilities. ChatGPT is highly favored because it provides users with services such as grammatical error correction and logical restructuring through text-based interaction, offering specific suggestions for improvement. Furthermore, the tool can intelligently recommend relevant literature and assist in resource retrieval based on the user's conversation history and current queries.

The GPT-4 model supports multimodal input, including text and images, and can provide instantaneous responses to voice queries with significantly improved processing speeds. Additionally, the model features a real-time web search function that overcomes the temporal limitations of its training database. Once users manually enable this option, they can access the latest information from 2024. Regarding file processing and code generation, it can directly read and analyze files such as Excel spreadsheets while supporting the writing and debugging of code in programming languages like Python and JavaScript. This is particularly beneficial for academic data processing, leading many scholars to utilize this technology for literature searches and to enhance the efficiency and quality of their paper writing.

ChatGPT and similar writing tools are increasingly well-suited for handling writing tasks within a Chinese linguistic context. In terms of content creation, these tools can engage in deep expansions of academic themes, integrating the latest research findings to output logically rigorous and well-argued paragraphs. Furthermore, they can assist in the drafting of abstracts.

引言

Highlighting research priorities and adhering to formal specifications, these tools offer reference formatting in various academic styles (such as Chicago) and can adjust manuscript layouts according to specific journal requirements. In terms of auxiliary refinement, they provide grammatical error correction to enhance the accuracy and fluency of expression, while also assisting in data analysis and literature interpretation to facilitate the efficient completion of academic writing tasks.

Furthermore, DeepSeek is an intelligent interactive system independently developed by the Chinese AI technology company DeepSeek. Since its launch, DeepSeek has rapidly garnered significant attention from both academia and industry due to its specialized capabilities in vertical domains. The latest upgraded version, DeepSeek-R1, introduces an academic enhancement suite that supports cross-linguistic intelligent literature retrieval, research hotspot analysis, and structured paper generation. Its built-in code interpreter can directly process scientific datasets from platforms like MATLAB and generate visual analysis reports. Notably, the system utilizes knowledge distillation technology to improve model inference efficiency, significantly reducing computational power consumption while maintaining high output quality. This characteristic has made it widely favored in university laboratories and industrial research institutes.

While these various systems each excel in different fields and provide users with a diverse range of options, it remains critically important when using these tools to uphold academic integrity and ensure that all generated content meets the standards of originality and authenticity.

2.2 人工智能对写作效率的提升

Exhibiting excellence in reducing writing time, the traditional writing process often requires a significant amount of time for data collection, conceptualization, drafting, and iterative revisions. The application of technology has made these steps more efficient. For instance, Semantic Scholar can quickly provide researchers with relevant academic papers and citations, substantially shortening the time required for literature collection. This efficient retrieval capability allows researchers to devote more time to analysis and writing. Furthermore, these tools offer significant advantages in improving grammatical accuracy and linguistic quality. For many authors, particularly non-native English speakers, the correct use of grammar and vocabulary remains a major challenge.

Grammar checking and language optimization tools, such as Grammarly and ProWritingAid, can automatically detect and correct grammatical errors, spelling mistakes, and stylistic issues. These tools also provide detailed explanations and suggestions to help authors understand the underlying reasons for their errors. In the context of academic writing, Turnitin Feedback Studio provides powerful auxiliary functions; it not only detects plagiarism but also offers targeted writing feedback to help graduate students and researchers improve the originality and quality of their papers. Additionally, the writing assistant tool Scrivener offers organization and planning features that help authors better manage large documents and complex writing projects.

However, these tools may still have deficiencies when processing complex logical relationships and deep professional knowledge. Therefore, authors must carefully review and edit AI-generated content to ensure the accuracy and reliability of the information. Overall, the improvement in writing efficiency provided by AI is evident. By saving time, enhancing linguistic quality, and providing targeted feedback, it has become an indispensable tool in modern writing. With the continuous development of technology, we can foresee that AI will play an even more important role in the future of writing, bringing more convenience and possibilities to authors.

Academic institutions both domestically and internationally have taken various stances regarding the participation of ChatGPT in paper writing. The impact of text-generating language models like ChatGPT on academic ethics has drawn significant attention from all sectors of society. Consequently, many internationally renowned academic institutions have issued relevant regulations. For example, the journal Science has stated that text, data, and charts directly generated by ChatGPT cannot be used directly, and authors must disclose the extent of the tool's involvement in the writing process. The domestic academic community has also paid extensive attention to the impact of ChatGPT on academic ethics. Several Chinese journals, including Chinese Clinical Oncology, have released regulations concerning the use of ChatGPT or other generative AI technologies during the writing and peer-review processes. Furthermore, Luan Yufan and others have analyzed the logic of ChatGPT from the perspective of the objectivity of academic ethical issues, identifying potential crises such as academic fraud, disputes over authorship and copyright, and the blurring of objectivity in talent evaluation. You Junzhe argued that the powerful capabilities demonstrated by generative AI, represented by ChatGPT, are transforming scientific research methods, leading to a conflict with existing research evaluation rules and increasing the risk of academic misconduct and fraud. Wang Youmei and others have further explored the developmental trajectory and connotative characteristics of ChatGPT from the perspective of game theory.

分析

The ethical risks associated with the application of ChatGPT in scientific research include the leakage and improper use of data privacy, issues inherent in machine algorithms, and the potential for dishonesty and imbalance in academic equity. It is evident that the intervention of ChatGPT in manuscript preparation exerts a multidimensional influence on academic ethics. Consequently, resolving the academic ethical crisis in the era of digital intelligence has become a pressing priority. Current domestic discourse reveals mixed evaluations regarding ChatGPT's role in paper writing; while the tool can significantly enhance research efficiency and help researchers expand their conceptual boundaries, its use remains controversial.

ChatGPT serves as a powerful auxiliary tool in stages such as literature recommendation, content creation, and manuscript revision. However, if researchers become overly dependent on such software, it may foster cognitive reliance, leading to potential risks such as the erosion of innovative capacity and the disordering of academic norms. Researchers should maintain a cautious attitude, prioritize their own academic abilities and professional expertise, and rationally weigh the benefits against the drawbacks.

To address the challenges posed by artificial intelligence to research integrity, we designed an experiment to test the authenticity and accuracy of ChatGPT as a writing tool. ChatGPT was prompted to compose an article on "how matrix stiffness affects the immunomodulatory capacity of mesenchymal stem cells." We also issued a separate command requiring the tool to provide references for the generated article and to conduct a critical analysis of them. The format of the content provided by ChatGPT is shown in [FIGURE:1]. While ChatGPT produced an article that appeared correct based on our instructions, we were unable to access the entirety of the references provided by ChatGPT through internet resources; these references were incorrectly listed in the article generated by ChatGPT.

Regarding the research paper generated by ChatGPT, we conducted a search for the titles of the references provided for the test paper using PubMed, Google Search, and the official websites of the cited journals. The results of these searches are detailed below.

A rigorous review was conducted on the references provided by ChatGPT for the test research paper. Among them, the reference Wang, (2020) was analyzed.

Matrix stiffness regulates immunomodulatory properties mesenchymal cells MCP-1 secretion.

A search for the title of this literature was conducted across Research Therapy, 11(1), PubMed, and the Google search engine, yet no results were found. Further investigation was carried out within the provided journal, Research Therapy; however, this publication is actually an international peer-reviewed journal focused on translational research in stem cell therapy. A search of the official BioMed Central journal website for the specific title also yielded no relevant records [FIGURE:N]. These findings indicate that the reference mentioned by ChatGPT does not exist.

参考文献

(2017). Matrix stiffness modulates effectiveness mesenchymal cells macrophage immunomodulation.

International Journal of Biology, 2017. ChatGPT indicates that this article was published in the International Journal of Biology, an international journal published by Hindawi focusing on cell biology. However, this journal was not included in the Chinese Academy of Sciences (CAS) JCR partitions as of the specified month. A search for this article using the PubMed search engine similarly yielded no results. A thorough search was conducted on the journal's official homepage and within the specific volume and issue where the article was allegedly published; however, as shown in [FIGURE:1], no such record was found. This further demonstrates that ChatGPT is fabricating references. These references are entirely erroneous, highlighting that AI programs like ChatGPT cannot replace the necessity for accuracy and reliability in academic citations. The fact that references generated by ChatGPT cannot be matched within academic databases suggests they are not derived from authentic scholarly works but are instead constructed by the program itself.

The results of searching for the reference titles were verified using PubMed, Google Search, and the provided official journal website. Subsequently, the content generated by ChatGPT in the previous experiment was saved and uploaded to GPTZero, a tool that claims to detect AI-generated text. GPTZero estimated the probability of the text being written by a human or a human-AI mix; however, despite the program's claims of accuracy, the results were inconsistent. Plagiarism detection was also performed on the ChatGPT content. The Turnitin detection report indicated that the text shared similarities with other sources, and a portion was identified as AI-generated content. Based on the aforementioned experiments:

We believe that the inappropriate use of ChatGPT in scientific writing will lead to the dissemination of misinformation. The underlying reasons for this practice may include: researchers choosing to use AI to quickly complete writing tasks due to pressing deadlines; researchers relying on AI tools to compensate for a lack of data or literature when research resources are insufficient; researchers sacrificing quality for the sake of publication volume due to various pressures; and a lack of awareness among some researchers regarding the ethical and legal consequences of the improper use of AI tools. These factors may result in researchers being unable to accurately distinguish between original text and AI-generated text or citations. [TABLE:1] presents the probability percentages of the test research papers being written by humans or a human-AI mix, and [FIGURE:2] provides the plagiarism detection report for the test research papers.

3 AI

Research Integrity Issues in Academic Writing and Their Solutions

Research Integrity Issues Arising in Paper Writing

Based on the aforementioned experiments, it is evident that generative language models such as ChatGPT can rapidly produce text that appears fluent and logically coherent during the academic writing process. While these tools enhance research efficiency, they also introduce new challenges regarding research integrity in academic writing.

First, the content generated by these models lacks true originality. Because these models rely on learning from and imitating vast amounts of existing data, the generated text is essentially a splicing and reorganization of pre-existing information. Consequently, when researchers rely on these tools for paper writing, they may unintentionally produce content that bears a high degree of similarity to existing literature—a practice that is frequently categorized as plagiarism.

Second, when generative tools produce content, they often cite academic resources or information sources. However, these citations are frequently inaccurate or fail to adhere to established academic standards. This can lead to the propagation of misinformation or the fabrication of references, further compromising the credibility of the scholarly work.

Academic Citation Standards and Scholarly Norms

The current system demonstrates a limited understanding of academic citation standards and scholarly norms. Consequently, the citations it generates may be inaccurate, improperly formatted, or entirely fabricated. In academic writing, adherence to rigorous citation standards is essential for maintaining intellectual integrity, allowing readers to trace the origins of ideas, and providing proper credit to original researchers.

When a model lacks a robust grasp of these conventions, several issues may arise. First, it may fail to distinguish between different citation styles, such as APA, MLA, or Chicago, leading to inconsistent formatting. Second, and more critically, it may produce "hallucinated" citations—references to papers, authors, or journals that do not exist. This poses a significant risk to the credibility of the work and violates the fundamental principles of academic honesty.

Furthermore, the system may struggle with the nuanced application of citations within the text. For instance, it might not correctly identify when a direct quote requires a page number or how to handle multiple authors according to specific stylistic guidelines. To ensure that a manuscript meets the requirements for peer-reviewed publication, all generated citations must be manually verified against primary sources to confirm their existence, relevance, and technical accuracy.

Fictitious or erroneous citations are a significant concern. Furthermore, generated text often lacks a deep understanding of academic literature; it may take fragments of information out of context, leading to the misunderstanding or distortion of originally accurate theories or perspectives. Consequently, based on the nature and severity of these issues, we categorize the research integrity problems arising from the use of AI tools in academic writing into several types: fictitious citations, erroneous citations, context stripping, inappropriate citations, "patchwork" generation, and the citation of unreliable sources. These issues not only undermine the rigor of academic research but also pose a serious threat to the foundational trust within the academic community.

Issues in Academic Citation and Research Integrity

The proliferation of generated content in academia has introduced several critical issues regarding citation integrity and research quality. These problems can be categorized into several distinct types of misconduct:

Fabricated Citations: This involves the inclusion of references that do not exist or are entirely fabricated, thereby misleading the reader. This category includes the use of "hallucinated" literature or the falsification of journal data.

Erroneous Citations: This occurs when an author cites the wrong document or misrepresents information within a source without proper attribution. In these cases, the research results may be accurate, but they are incorrectly attributed to a specific reference.

Contextual Distortion: This issue arises when fragments are extracted from a text without considering the surrounding context, leading to a misinterpretation of the original author's intent. This often involves quoting specific passages in a way that misleads the reader about the source's actual findings.

Inappropriate Citations: This refers to the practice of citing sources that have weak relevance to the topic or lack academic value. Examples include citing non-academic literature or failing to provide an appropriate scholarly background for the claims being made.

Patchwork Generation: This involves the "stitching together" of content from multiple documents without proper synthesis or the extraction of original insights. Instead of conducting an original analysis, the author merely assembles fragments of existing literature into a cohesive-looking but unoriginal whole.

Reliance on Unreliable Sources: This occurs when information is drawn from untrustworthy sources or those whose credibility is in question. Citing low-quality journals or articles that have not undergone rigorous peer review significantly undermines the overall quality and authority of the research paper.

Limitations of Current Oversight Mechanisms

Current mechanisms for supervising scientific research integrity face significant deficiencies. Existing academic detection tools struggle to identify the originality of generated content; most plagiarism detection systems rely primarily on comparisons against established databases of existing literature and lack effective means to identify synthetically generated text.

Furthermore, academic journals and research institutions lack robust screening mechanisms during the review process to fully identify potential academic misconduct associated with generated content. these shortcomings have created a regulatory "blind spot" in the application of new technologies within the academic community, further exacerbating academic risks.

To address these challenges, the academic community must strengthen the formulation of standards for the use of digital tools. It is essential to establish more comprehensive research integrity supervision mechanisms and increase the rigor of the review process for generated content to effectively prevent potential academic misconduct.

Strengthening Ethical Awareness Education for Researchers Regarding AI Tool Usage

The widespread adoption of AI-assisted writing tools has played a significant role in improving work efficiency and optimizing research outcomes. However, it is increasingly critical to strengthen ethical awareness education for researchers regarding the use of these tools. First, researchers should enhance their ethical understanding of AI usage through systematic training and dedicated coursework. For instance, AI-generated content may inadvertently infringe upon copyrights, or generated citations may be inaccurate, potentially leading to plagiarism and academic misconduct. Training programs should emphasize these potential issues, teaching researchers how to identify and avoid such ethical pitfalls. By reinforcing ethical awareness, researchers can exercise greater caution when utilizing AI tools and prevent improper behavior.

Furthermore, researchers must proactively adhere to ethical norms when employing AI-assisted writing. For example, when using AI to generate content or citations, researchers should explicitly disclose the extent of AI involvement and accurately indicate which sections of the manuscript were produced with the assistance of AI tools. At the same time, researchers must maintain a critical perspective toward AI-generated content. They should not blindly rely on the results provided by these tools; instead, they must integrate their own academic judgment to proofread and revise the material, ensuring that the final content meets rigorous academic standards.

In the educational process, case studies, group discussions, and practical exercises should be utilized to help researchers understand the scope of application and the inherent limitations of AI technology. This approach guides them to remain vigilant when using these tools, preventing academic integrity issues caused by over-reliance or misuse. Through comprehensive ethical education, the research community can harness the benefits of AI while maintaining the highest standards of scholarly integrity.

Furthermore, the misuse of such tools may lead to a decline in the quality of academic work. Finally, research institutions should provide researchers with clearer ethical guidance and policy support. Research ethics committees and academic journal editorial boards must keep pace with rapid technological developments by issuing relevant policies and guidelines that explicitly define the norms for using these tools in scientific writing.

At the same time, institutions should regularly organize ethics lectures and seminars, inviting experts and scholars to share insights on the ethical issues and solutions associated with the use of these technologies. Such initiatives will enhance the sense of moral responsibility among researchers. Through comprehensive ethical education and training, researchers will be better equipped to understand and address the challenges posed by these technologies, thereby elevating the overall standards of academic integrity.

Enhancing Supervisory Capabilities Through Technological Means

In the contemporary academic environment, the continuous development of technological tools provides critical support for enhancing supervisory capabilities, particularly through systems based on machine learning for plagiarism detection and the verification of citation accuracy. These technologies not only improve the efficiency of oversight but also offer robust safeguards for maintaining academic integrity. Plagiarism detection tools, in particular, play a vital role in the processes of academic writing and scholarly publishing.

Turnitin is currently one of the most widely utilized plagiarism detection tools. It functions by comparing submitted manuscripts against an extensive database of existing texts to identify textual similarities and generate comprehensive similarity reports. The underlying mechanism of this tool relies on sophisticated text-matching technology, which enables the precise identification of plagiarized content, ranging from direct verbatim copying to more subtle paraphrased or modified material.

Turnitin is capable of detecting not only traditional plagiarism but also the subtle differences between AI-generated content and human-authored text by analyzing structural patterns, content redundancy, and shifts in sentiment. Beyond distinguishing between machine and human writing, it utilizes semantic analysis to evaluate the originality and contextual understanding of the content, further determining its authenticity.

In addition to plagiarism detection, addressing the issue of "hallucinated" or entirely erroneous citations is a critical challenge. To mitigate this problem, specialized algorithms must be developed and implemented to detect these false references. The specific technical roadmap for this includes the following:

• Citation Consistency Checks: Developed algorithms should interface with major academic databases (such as Google Scholar, PubMed, etc.) to automatically verify the authenticity of cited works. This ensures that the referenced literature actually exists in the database and corresponds accurately to the cited content.
• Literature Content Analysis: Algorithms must analyze the actual content of the cited literature to ensure that the themes and findings of the reference align with the arguments presented in the paper. This prevents AI from randomly splicing irrelevant citations into generated text.
• Contextual Understanding: It is necessary to enhance the contextual understanding capabilities of these algorithms to avoid the phenomenon of quoting out of context. In some cases, a source may be cited to support a specific viewpoint, even though the original intent of the source does not fully align with the author's claims. Strengthening context-aware capabilities ensures that cited literature accurately reflects its original meaning.

Establishing automated review mechanisms is essential for ensuring the accuracy of citations. By regularly checking and validating the authenticity of generated content, researchers can promptly identify potential errors or false citations, thereby safeguarding the overall quality and integrity of academic work.

Establishing Transparent Disclosure Mechanisms and Promoting Academic Rule Updates

Zheng Wenwen and colleagues from the Institute of Scientific and Technical Information of China (ISTIC) note that various publishers and institutions have already formulated usage policies regarding AI-generated content for authors, editors, and reviewers. Currently, the requirements for authors regarding the use of AI in the academic publishing process are relatively specific and cover a comprehensive range of content. However, specific regulations remain inconsistent across organizations. Furthermore, there are fewer requirements governing the use of AI technologies by editors and reviewers, with some publishers and institutions failing to provide any concrete guidelines at all.

To address these challenges, the academic community urgently needs to establish transparent disclosure mechanisms and promote the updating of relevant rules to ensure the compliance of AI technology in scientific research. First and foremost, establishing a transparent disclosure mechanism is a fundamental pillar for ensuring research integrity. To prevent behaviors such as over-reliance on AI, plagiarism, or improper citation, academic institutions should require researchers to explicitly declare the use of AI tools and their specific purposes within their papers. For example, researchers could specify the following in the methods section or acknowledgments:

This study utilized the OpenAI ChatGPT tool to assist in generating the preliminary draft of the paper. During the analytical phase, specialized data analysis tools were employed. This disclosure mechanism not only enhances the transparency of the research but also provides peer reviewers and readers with a clearer understanding of the technical methodologies and tools the authors relied upon throughout the research process.

Furthermore, publishing institutions can implement a series of measures to ensure compliance in the use of these tools. For instance, journals may require submitted manuscripts to adhere to strict standards regarding the citation and generation of content, establishing specialized review mechanisms to evaluate whether the use of such tools aligns with academic ethical requirements. If it is discovered that AI-generated sections are not correctly labeled, publishers may require authors to make corrections or even reject the paper for publication. Such measures not only promote research integrity but also prevent the potential abuse of these technologies within the academic community.

Finally, schools and educational institutions should play an active role in this process. Universities can integrate guidelines for the use of AI tools into their curricula to help graduate students understand the appropriate ways to utilize these technologies in scientific research. Simultaneously, educational institutions should issue formal statements regarding their stance on the use of such tools and explicitly define the requirements for labeling generated content within their academic codes of conduct. This ensures that students adhere to these principles throughout their future academic careers.

4 结论

While the integration of artificial intelligence (AI) undoubtedly plays a positive role in enhancing the efficiency and quality of academic writing, its misuse has led to numerous instances of research misconduct, such as academic dishonesty and ethical violations. These issues not only hinder the development of the academic publishing field but also disrupt the orderly conduct of scientific research. Therefore, as we promote the widespread application of AI technology in academic writing, we must remain cautious regarding its potential negative impacts. This paper summarizes the research integrity issues arising from the use of AI in writing and proposes several recommendations. First, the academic community should strengthen the supervision of AI tools in academic writing by establishing clear ethical guidelines and codes of conduct. Second, scholars must maintain a high level of academic integrity when using these tools, ensuring the originality of their content and the accuracy of their citations. Simultaneously, academic publishing platforms and journals should intensify the research and development of AI detection technologies to promptly identify potential misconduct. However, this study has certain limitations, primarily reflected in the limited scope of data coverage and an insufficient sample size. Future research could further explore more advanced content detection methods and enhance the development of automated auditing technologies for AI-generated content to ensure the authenticity and credibility of academic achievements. Furthermore, in-depth investigations are needed to determine how to balance the relationship between AI technology and academic ethics, ensuring the coordinated development of technological progress and research integrity.

References: MALIK, PRATIWI, ANDAJANI. Exploring Artificial Intelligence in Academic Essays:

Higher Education Student's Perspective International Journal Educational Research Open, 2023, ALMUSHARRAF ALOTAIBI error-analysis study writing context:

Human Automated Essay Scoring Approaches Technology, Knowledge Learning, 2022, MIRANTY WIDIATI automated writing evaluation (AWE) higher education Pegem Journal Education Instruction, FARROKHNIA BANIHASHEM NOROOZI

analysis

ChatGPT: Implications educational practice research Innovations Education Teaching International, 2024, 61(3):

ROSPIGLIOSI Artificial intelligence teaching learning: questions should ChatGPT?

Interactive Learning Environments, 2023, 31(1):

JOVSK KASAL Artificial Intelligence Generate Fraudulent Authentic-Looking Scientific Medical Articles:

Pandora's Opened Internet 2023, e46924. ZIMMERMAN Ghostwriter Masses:

ChatGPT Future Writing Annals Surgical Oncology, 2023, 30(6):

China Academy of Information and Communications Technology & JD Explore Academy. AI-Generated Content. China Information Technology. Response Strategies of Scientific Journals to the Impact of ChatGPT. 2024, (01).

Opportunities, Challenges, and Strategies for Assisting Social Science Graduate Students in Research Writing. DeepSeek. Journal of Chongqing University of Technology (Social Sciences Edition). THORP. ChatGPT. Science, 2023, 379(6630). Chinese Journal of Oncology: Regulations Regarding the Use of ChatGPT or Other Generative Artificial Intelligence Technologies in Paper Writing and Peer Review. 2024, 33(09). Shanghai Nursing: Regulations Regarding the Use of Generative Artificial Intelligence Technologies in Paper Writing and Peer Review. 2025, 25(04).

Research on Potential Risks to Academic Ethics and Response Strategies Arising from the Rise of ChatGPT. Reform and Opening. 2023, (06). Application Risks and Control Measures of ChatGPT-like Generative Artificial Intelligence in Research Scenarios. Information Theory and Practice. 2023, 46(06).

Ethical Risks and Avoidance Paths of ChatGPT in Educational Applications. Open Education Research. 2023, 29(02). Research on Constructing a Framework for the Use of AI-Generated Content in Academic Publishing. Chinese Journal of Scientific and Technical Periodicals. 2024, 35(10).

E-mail: Proposed the research ideas and designed the research protocol.

实验

Author Contribution Statement

Tenzin Lhamo and Yidi Sun were responsible for drafting the manuscript and conducting data collection and analysis. All authors participated in the revision of the final version of the paper.