Artificial Intelligence in Manuscript Preparation: Are We Becoming Dependent on Machines?

Sir,

We read two interesting letters – one discussing about the usage of a large language model (LLM) chatbot in academic writing[1] and another discussing the challenges of using such tools in academic writing.[2] I agree with both the authors that LLM like ChatGPT has potential in academic writing and there are challenges in using it.

For manuscript preparation, authors can potentially use three routes [Figure 1]. They can write their manuscript which is the best practice. Many researchers are so overburdened with other works that they take services from other agencies who generate (analyze, interpret, and write) manuscripts. This manuscript is then edited by researchers and the agency is acknowledged. The agency charges a hefty amount for the service. Generative artificial intelligence (AI) chatbots can be used to generate (analyze to some extent, interpret, and write) manuscripts. Authors can provide suitable prompts to generate texts and can edit them to prepare a coherent manuscript. Similar to acknowledging agency, the authors need to acknowledge the usage of AI tool.[3]

Close

Figure 1 Three ways of manuscript preparation. AI: Artificial intelligence

Export to PPT

The current guidelines by the International Committee of Medical Journal Editors (ICMJE) and the World Association of Medical Editors (WAME) suggest that AI tools can be used in manuscript preparation with human oversight and proper acknowledgment.[4] However, a few of the journals and book publishers are barring the publication of so-called AI-generated text. In this context, I would like to share three of my experiences. My fellowship guide informed me that he would not renew his subscription of a popular grammar correction software as AI detectors detect AI-generated text when manuscript is edited by the software. In another instance, we prepared a chapter for an edited book. We used an LLM for correcting grammar and polishing the language. When we sent the chapter, the editor sent it back immediately citing that AI-generated content should be below 10%. In a similar case, although the editor approved the chapter, the production house returned the chapter due to high AI-generated content. However, one coauthor paraphrased the chapter using an AI tool and it was cleared for publication.

From the example above, it becomes evident that some of the editors or publication houses are using AI-detector without actually reading the content. Moreover, if AI tools suggest that the manuscript does not have AI-generated content, it is accepted. This practice may be similar when a journal editor or technical staff screens a manuscript for text similarity first, then only processes it further. Hence, the “text” rather than the “content” takes the upper hand in publication.

Due to this practice of few publishers or editors, the industry is taking its way of influencing authors. Figure 2 presents such an example. AI chatbot generates or edits text, AI detectors detect AI-generated content, and AI tools paraphrase or edit it to bypass the AI detectors. However, if we have a fair knowledge about ICMJE and WAME guidelines, the lower three steps [Figure 2] could be avoided.

Close

Figure 2 Usage of artificial intelligence in academic writing. AI: Artificial intelligence

Export to PPT

Scientific language needs clarity to convey the findings. In this process, if AI helps, it is no harm to science. Hence, both ICMJE and WAME have not barred the use of AI in academic writing. Both of them have clear guidelines for the transparent and responsible use of the tools. I urge stakeholders to disseminate those guidelines so that “AI phobia” in academic writing is reduced among the authors. Authors need to remember a thumb rule while using AI in academic writing – use, edit, take responsibility, and acknowledge![5]