Public health journals are setting guidelines for use of AI in submissions as the technology becomes commonplace.
Photo by Boy Wirat, courtesy iStockphoto
With a manuscript on artificial intelligence in the works, researcher Richard Bruno, MD, MPH, is not observing the debate over the use of AI in journal publishing from the sidelines. He is in the middle of it.
A physician, health officer, researcher and early AI adopter in Portland, Oregon, Bruno is actively working to understand how AI tools can help public health publishing without crossing ethical lines. Recently, he spoke at an AI user's group in Portland about the need for guardrails around privacy, copyright, bias, workforce impacts and environmental costs.
“AI can definitely be helpful, but we have to use it judiciously,” Bruno, a family and preventive medicine physician and health officer at the Multnomah County Health Department, told The Nation's Health.
The rapid rise of generative AI has forced public health journals and editors to confront a question that is no longer theoretical: Can researchers use AI to improve scientific publishing without weakening the integrity of science?
The answer from journal editors and publishing leaders is a cautious “yes.” AI can be used ethically in public health publishing, they say, but only when researchers are transparent about how they use it and remain fully accountable for the work.
In recent interviews with The Nation's Health, publishing leaders and editors emphasized the need for clearer standards across journals, careful disclosure and continued human oversight. AI may help with writing, editing, literature review or parts of a study, they said, but it cannot replace original thought, scientific judgment or responsibility for the final product.
Learning how to constrain AI and write effective prompts, how to properly review and vet generative information, and how not to slip into passive thinking is paramount, according to Brian Selzer, APHA's director of publications services.
“At all stages of writing, evaluation and publishing, papers should be human generated first and then refined by AI, followed by re-review by a human actor,” Selzer told The Nation's Health.
Across public health and medical studies, AI is being used for a growing list of tasks, such as improving grammar and readability, helping non-native English speakers communicate findings, organizing references, summarizing literature, checking formatting, identifying keywords and assisting with data coding or visualization. Journals and researchers may also use AI tools to flag plagiarism, duplicate submissions, image manipulation or possible paper mill activity.
There are many examples of responsible and transparent use of AI in research, and the rules have grown clearer in recent years, according to Annette Flanagin, RN, MA, executive managing editor and vice president of editorial operations for the Journal of the American Medical Association and the JAMA Network.
JAMA has been publishing about the use of AI in medicine for years, Flanagin said. For example, a 2016 JAMA study on using deep-learning algorithms to detect diabetic retinopathy from eye images showed AI's promise, Flanagin said. But it also raised concerns that remain relevant today — whether models are tested on representative populations, whether they are validated before use and whether clinicians and researchers can trust them.
And there are firm limits. Echoing the concerns of her counterparts, Flanagin stressed that AI tools cannot be listed as authors and should not be used in ways that violate confidentiality, intellectual property rights or journal disclosure rules.
Additional risks include fabricated or erroneous content, confidential patient or participant information being entered into public models and intellectual property being stored or used to train AI systems. As such, journals should test AI tools carefully before relying on them, Flanagin said.
“We encourage piloting, testing and validating AI tools in the editorial process, learning what works and what does not, and prioritizing the principles of transparency, responsibility and accountability in all actions,” Flanagin told The Nation's Health.
But the editors who see promise in AI also draw clear boundaries around its use. Authors, they said, must disclose meaningful AI assistance, verify AI-generated material and remain fully responsible for accuracy, interpretation, confidentiality and scientific integrity.
“There is consensus on several fundamental issues, such as that an AI tool cannot be an author on a paper,” said Magdalena Skipper, PhD, editor-in-chief of Nature. “The reason for it is very straightforward — with authorship comes accountability, as well as credit. Accountability always has to remain with the researcher.”
As AI tools evolve, journal editors and publishing leaders are moving quickly to update guidance and train researchers on responsible use. Nature and other journals in the Nature Portfolio have public instructions for authors and reviewers, Skipper said, and those instructions will continue to change as technology does.
“As the AI tools develop, of course we will be revising our guidelines, just as we do with all other guidelines,” she said.
JAMA has also published guidelines. A March 2024 editorial provided authors, researchers, reviewers and editors with information to help guide the transparent and accountable reporting of AI's use in research and publication, Flanagin said.
Similarly, APHA's American Journal of Public Health will soon publish its own policy on AI use, said AJPH Editor-in-Chief Denys Lau, PhD.
“Researchers can use AI ethically provided they fully disclose how they used AI in the preparation of their submission to the journal,” Lau told The Nation's Health.
AJPH also does not allow AI to be listed as an author of submissions, Lau said, noting that researchers are expected to take full accountability of the content and write-up in their content. Moreover, AI should not be used to generate an author's original ideas, conceptualize a research question or replace the unique personal voices or creative thoughts of the researchers.
“Public trust and trustworthiness in public health science have declined in recent years,” Lau said. “The burden falls on every public health researcher and professional — including AJPH editors, authors and peer reviewers — to continue to uphold the scientific integrity of peer-reviewed science and not let AI-generated materials erode the established credibility of scholarly publications.”
But protecting that credibility is becoming harder as AI tools move deeper into the mechanics of research and publishing, leaders say.
Margaret Nolan, MD, MS, deputy editor of the American Journal of Preventive Medicine, said she sees clear benefits in AI, especially when it saves researchers and editors time on formatting, word counts, copy editing, English-language support and finding peer reviewers. But at the same time, she worries that AI may crowd out original thought, making it easier to generate manuscripts.
Nolan said she is concerned about a rise in incremental research — studies that build only slightly on existing work, often using the same datasets with a new variable or a different comparison. Such papers may be technically sound, she said, but still fail to answer the most important question for editors: Why is this important?
Academia bears some responsibility because publishing remains central to career advancement, creating pressure to produce more papers. But making manuscript writing faster does not necessarily mean scientific or public health progress will come faster, she said.
Instead, editors may spend more time sorting through submissions to identify work that matters and separating it from those that are “papers to be papers.”
That issue is especially troubling to Jagdish Khubchandani, PhD, MPH, MBBS, a professor of public health sciences at New Mexico State University and secretary of the World Association of Medical Editors. Generative AI has intensified longstanding concerns about so-called “paper mills,” defined as businesses or networks that produce low-quality or fake manuscripts for authors under pressure to publish.
Khubchandani said editors are seeing more warning signs: authors with impossibly high publication counts, papers with excessively polished language and batches of similar articles using familiar topics or datasets. Some papers may be rejected by one journal, he said, only to be sent elsewhere.
He compared the problem to a wildfire, fueled by pressure to publish, weak enforcement and AI tools that can generate polished manuscripts in minutes.
Like other publishing leaders, however, Khubchandani sees AI as useful when it supports human work, such as improving language, helping researchers frame ideas more clearly or speeding evidence synthesis.
That distinction is important to Justin Moore, PhD, MS, editor-in-chief of the Journal of Public Health Management and Practice.
For Moore, the more realistic path to finding consensus in the AI debate is to accept that AI is already part of scientific publishing and focus on helping researchers and journals use it responsibly.
“This fretting and worrying, I just don't see it,” Moore told The Nation's Health. “What we need to do is put up guardrails. We need to put out guidance. We need to put out education.”
Back in Portland, Bruno and his co-authors are working within that reality as they develop their manuscript on AI and public health emergency preparedness. He sees a need for researchers, editors, reviewers and public health practitioners to work together to define responsible use, rather than leave authors to navigate the technology alone.
“It's really important to get this right out of the gate,” Bruno said.
For more information, visit bit.ly/jamaguides and bit.ly/natureguidance.
- Copyright The Nation’s Health, American Public Health Association
