To address the increasing use—and misuse—of generative artificial intelligence (AI) tools based on large language models (LLMs) in medical publishing, organizations that set standards for publishing ethics, such as the International Committee of Medical Journal Editors (ICMJE), the World Association of Medical Editors (WAME), and the Committee on Publication Ethics (COPE), have issued updated guidelines and position statements on the responsible use of AI.
In parallel, leading journals such as The Lancet, The New England Journal of Medicine, Nature, and those in the JAMA Network have laid down ground rules on acceptable AI use and appropriate disclosure. Many other journals have followed suit, updating their instructions for authors to reflect their own policies. For instance, the Japan Society for Medical Education (JSME) recently revised the author guidelines for its journal Medical Education (Japan) to include a new section, 7.3 Use of Artificial Intelligence (AI)-Assisted Tools/Technologies.
With this wealth of new guidance, one might expect that authors now have a clear understanding of best practices regarding AI use. However, that is far from the reality. A recent Nature survey of 5,000 researchers worldwide found that opinions on the acceptable use of generative AI in manuscript preparation remain deeply divided. Many researchers are uncertain about what constitutes responsible AI use, and whether disclosure is even necessary.
The above guidelines all share a common theme: transparency, accountability, and the human responsibility to ensure scientific accuracy. They all agree that AI tools do not qualify for authorship. They do differ, however, in the extent and manner of disclosure that is required when AI tools are used.
So, how should the use of AI be disclosed?
Nature in its Instructions for Authors states that “Use of an LLM should be properly documented in the Methods section (…) of the manuscript.” There is no mention, however, of what constitutes the “use of an LLM,” leaving this open to wide-ranging interpretation.
The revised WAME Recommendations on Chatbots and Generative Artificial Intelligence in Relation to Scholarly Publications, acknowledge that researchers may use AI tools to perform different functions: “1) simple word-processing tasks (analogous to, and an extension of, word-processing and grammar-checking software), 2) the generation of ideas and text, and 3) substantive research.” Based on this, WAME recommends that use of an AI tool “to draft new text” (i.e. function 2), should be noted in the Acknowledgment section, and use of an AI tool to “carry out or generate analytical work, help report results (e.g., generating tables or figures), or write computer codes” (i.e. function 3), should be stated in the body of the paper, in both the Abstract and the Methods section.” There is no recommendation regarding the use of an AI tool for function 1, suggesting that this does not warrant disclosure.
JAMA in its Instructions for Authors makes the same distinction between AI used in manuscript preparation and AI used in research, the former requiring disclosure in the Acknowledgment section and the latter in the Methods section. JAMA also states explicitly that disclosure is not needed for the use of AI as a basic tool for checking grammar, spelling, references, etc.
In the May 2023 update of its Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals, the ICMJE added a new section, 4. Artificial Intelligence (AI)-Assisted Technology, including the rather vague guidance that “Authors who use such technology should describe, in both the cover letter and the submitted work, how they used it.” The subsequent January 2024 update clarified that use of AI for writing assistance should be reported in the Acknowledgment section, while use of AI for data collection, analysis, or figure generation should be described in the Methods section.
As the technology keeps evolving, these guidelines are sure to be revised over the coming years. For now, however, the following is my take on good practice regarding AI disclosure:
- If AI was used as part of the research methodology (and therefore included in the research plan), then describe this in the Methods section along with sufficient details, including prompts, name and version of the software, etc., to allow other researchers to reproduce the results (as you would do for other tools such as statistics software.)
- If AI was used to newly generate text in the process of creating the manuscript, provide details in the Acknowledgment section.
- If AI was used to translate the text written by the authors, state this in the Acknowledgment section (as is customary when a human translator is used.)
- If AI was used to substantively edit the text written by the authors, similar to how it would be extensively revised by a professional medical English editor, then state this in the Acknowledgment section (as is customary when a human editor is used.)
- If AI was used to do basic editing of the text written by the authors, for checking grammar, spelling, references, etc., then no disclosure is necessary.
The difference between 4 and 5 may be a subtle one. If you feel that, after correction and polishing by AI, the resulting text remains your original wording, then disclosure is probably not necessary. If, however, the text is revised by AI to the extent that you feel it is no longer your original wording, then disclosure is necessary. If you’re not sure, disclosure is probably a good idea.
Of course, as with any recommendations regarding the preparation of manuscripts, always follow the Instructions for Authors of your target journal, and when in doubt, err on the safe side and consult the editor.