The Artificial Intelligence Dilemma in Academic Writing: Balancing Efficiency and Integrity

INTRODUCTION

The landscape of academic writing is undergoing a profound transformation with the advent of artificial intelligence (AI) tools, particularly large language models (LLMs) such as ChatGPT.^[1] These technologies have made it significantly easier for researchers, students, and professionals to draft and refine scholarly content.^[2] Especially for non-native English speakers or those facing difficulties with academic expression, AI offers assistance in producing grammatically correct and coherent texts.^[3] However, this convenience comes with growing concerns about the authenticity, originality, and intellectual ownership of AI-assisted writing.

In response to the increasing prevalence of AI-generated content, a new category of software – “AI detectors” has emerged. These tools aim to identify text that may have been produced by AI.^[4] The presence of such detection tools has, in turn, prompted the rise of yet another innovation - AI “humanizers” or paraphrasing tools designed to modify AI-generated content so that it escapes detection.^[5] This loop between generation, detection, and obfuscation has created a complex ecosystem that challenges traditional notions of academic integrity.

Although fully relying on AI-generated content can diminish the author’s credibility,^[6] we presume that its use would help the authors a lot in research and writing.^[7] In this context, we would like to share our perspectives on how the current growth in the AI industry is influencing authors and academicians and how to find a way forward.

HOW AI HELPS IN WRITING

AI can help in three stages of academic writing, as shown in Figure 1. It starts with an AI text generator, followed by an AI detector, and an AI paraphraser or so-called “humanizer.”^[8] While LLMs like ChatGPT can help in generating texts, tools like Turnitin can help detecting the AI content in the text, and tools like Stealth Writer can twist language to bypass the AI detection. With the easy availability of these three types of tools, academicians, authors, and researchers are becoming more confused and having a dilemma – “to use or not to use AI?” In the following section, we provide a brief of the three tools and then comment on the current guidelines for using AI for academic purposes.

Close

Figure 1:

Steps of the writing process where artificial intelligence can be used along with two examples of such tool. (AI: Artificial intelligence)

Export to PPT

THREE AI TOOLS

Detection

AI detection tools have been developed to identify content created by AI text generators like ChatGPT, Claude, or Perplexity. AI detectors analyze text for patterns and features that suggest the involvement of AI in the writing process.^[13] Some of the teachers and book editors started using it to screen submitted text.^[14] The problem with such a detector is that it will detect as “AI-generated” even if the language and grammar are edited by an LLM. For example, a part of a manuscript was copied from a previously published article.^[15] It was detected as 100% human-generated text. Then, it was edited for grammar and language by an LLM chatbot. Checking it by an AI-detector showed 100% AI-generated [Figure 2]. The scientific content of those two texts was the same. The report says one is human-generated, and another one is AI-generated. However, it was not AI-generated; it was just edited by AI.

Close

Figure 2:

An example showing how human written texts can pass through various artificial intelligence tools for detecting human-generated content, editing text, detecting artificial intelligence (AI)-generated text, “humanize” text, and bypassing AI detection, from top to bottom.

Export to PPT

In addition to this, these tools can also flag texts as “AI-generated and AI-refined,” and “Human-written and AI-refined.” However, in all cases, scientific content remains the same. Only the language is different.^[16] The same content with a different writing style is flagging the content as AI-generated. Hence, these tools are nothing but adding another level of unnecessary screening for scientific papers. We need to read science, not to investigate what tool was used to write it.

PROBLEM

If the journals, editors, and reviewers start detecting AI content in manuscripts and reject manuscripts on the basis of AI-generated content, authors may develop an “AI phobia,” and they would avoid using AI tools. A recent survey found divergent views of researchers on AI’s role in academic writing. Some argued that AI, like calculators, has become so common that disclosure should be unnecessary, reflecting a pragmatic normalization of its use. Others opined that using AI in writing or reviewing to be considered “cheating.”^[18] By rejecting AI, this group may struggle to stay competitive, as AI can assist with data analysis, literature review, and drafting, all of which can enhance research productivity.^[19]

On the other hand, another group of authors may edit or generate content and “humanize” the text, which may reduce the fluency and clarity of scientific communication. Over-editing can result in convoluted phrasing, loss of coherence, and a lack of smooth transitions.^[20,21]

Hence, some authors are in a dilemma, and some are using two tools (AI text generator and humanizer). A study by Ng et al.^[22] found that researchers think they need institutional support and formal training or clear policies on AI tool usage. The findings underscore a clear need for improved guidance, transparency, and educational initiatives around AI in research.^[22]

ROADMAP FOR POTENTIAL SOLUTION

LLM chatbots can be used in various stages of research and manuscript preparation, as shown in Table 1.^[11,23-25]

When using chatbots like ChatGPT in a research and academic writing, several precautions are essential to ensure ethical integrity, accuracy, and responsible use.^[26] Researchers must recognize that chatbots are tools, not substitutes for human expertise.^[27] While these AI systems can assist in drafting text, suggesting ideas, or summarizing content, their outputs should always be critically evaluated.^[28] Chatbots may generate plausible-sounding but factually incorrect or misleading information.^[29] Hence, researchers must cross-verify all chatbot-generated content with reliable primary sources, especially when citing literature.^[30]

Transparency is another crucial precaution. Any substantial use of chatbots in manuscript preparation, such as drafting, editing, or language enhancement, should be clearly disclosed in the acknowledgments or methods section.^[31] This fosters transparency and allows peer reviewers and readers to understand the extent of AI involvement in the work. For more detailed current guidelines, authors may visit the International Committee of Medical Journal Editors (ICMJE)^[32] criteria or World Association of Medical Editors (WAME) guidelines.^[33]

While using chatbots, confidentiality, and data privacy must be preserved. Researchers should avoid inputting sensitive, unpublished, or personally identifiable information into chatbot interfaces, as the terms of service of some AI platforms may allow usage data to be stored or used for training.^[34] This is especially critical in medical and clinical research, where patient confidentiality is a legal and ethical mandate.^[35]

Finally, the use of chatbots should not replace genuine scholarly engagement or critical thinking. While AI can help streamline some aspects of writing, the responsibility for the intellectual content, argumentation, and conclusions lies solely with the human authors.^[36] Both ICMJE and WAME emphasize that authors must take full responsibility for the accuracy and integrity of their work after thoroughly checking the content. A summary of the guidelines are shown in Figure 3.

Close

Figure 3:

Major guidelines for using artificial intelligence in academic writing.

Export to PPT

In addition, biomedical journals are also not against using AI in manuscript preparation.^[37] However, researchers should also ensure they are not violating the editorial policies of the particular journal, some of which may have specific rules about AI-generated content. For example, journals published by Scientific Scholars have a field where authors need to declare the use of AI in addition to acknowledgment.

In this era of AI-assisted writing and publication process, AI-detectors and AI-humanizers should be obsolete in the near future if authors start following guidelines for the optimum use of chatbots.^[38] Journal and book editors and reviewers should evaluate the scientific merit of the paper and not to check the AI report, except for extreme suspicion of unscientific argument.

PROTECTING WOMEN’S HEALTH DATA

Sharing women’s health data with chatbots raises particular concerns about privacy, confidentiality, and the ethical handling of sensitive information. Women’s health data often includes details on reproductive health, mental health, and other personal health factors that require a high level of confidentiality.^[39]

In the context of using chatbots for research writing on women’s health, researchers need to be cautious. They should verify each chatbot’s data handling policies to ensure that sensitive information is not stored or reused. When the chatbot stores or uses the input data for training models, it is suggested that the researchers use anonymized data in a prompt. This can provide an additional layer of protection.^[40]

CONCLUSION

Human effort is still the best for scientific writing; however, AI can augment that! The integration of AI into academic writing presents both opportunities and challenges, marked by the dynamic interplay between AI-generated content, detection tools, and humanizing technologies. The escalating use of AI necessitates a clear understanding of its capabilities and limitations, alongside a commitment to transparency and academic integrity. Adhering to established guidelines from the ICMJE and WAME, authors can ethically embrace AI with transparency. The academic community must focus on the judicious use of AI to augment productivity, keeping a vigilant eye on the scientific vigor of the scholarly work.