The write algorithm: promoting responsible artificial intelligence usage and accountability in academic writing
BMC Medicine volume 21, Article number: 334 (2023)
Recent strides in large language models, powered by sophisticated artificial intelligence (AI) algorithms trained on extensive language datasets, have revolutionised writing tools . OpenAI’s ChatGPT, a leading example, excels at analysing text and generating content based on user input. These breakthroughs have profound implications for academic writing, attracting the attention of journals worldwide . While the pros and cons of adopting these technologies have been extensively debated, the responsible implementation and transparent documentation of their use remain relatively overlooked. This Editorial seeks to fill this gap.
Potential of large language models in academic writing
Large language models offer immense potential in academic writing. They effectively process vast amounts of data, generating innovative ideas and even entire scholarly manuscripts. Through emulating human writing styles, they have the potential to enhance precision, authority, and emotional connection. These tools improve writing efficiency, allowing more focus on data analysis, and promote inclusivity by assisting authors with language barriers and providing translations for a wider audience. Integrating AI into academic writing can advance knowledge and foster a more interconnected and collaborative academic landscape, but ethical considerations are essential.
Recognising the current limitations of large language models
Acknowledging the limitations of large language models in human-like authorship is vital. These models rely on data compression, resulting in estimations and compilations rather than precise reproductions, which can lead to the generation of fabricated (“hallucinated”) data . They may also struggle to capture nuanced and specialised knowledge essential for accurate academic writing, similar to how foreign language learners misinterpret idiomatic expressions. It is worth noting that the vast databases large language models are trained on include text based on falsehoods and other sources of untruth . The inability to update training data in real-time also poses concerns regarding the potential inclusion of outdated information , particularly in rapidly evolving fields. Hence, exercising caution when considering outputs from these models is crucial, as they may encompass inaccuracies, fabrications (e.g. erroneous references), and potential instances of plagiarism . Ethical concerns are paramount in AI content generation . Large language models may inadvertently perpetuate biases, mirroring skewed representations of certain topics or populations . Failing to evaluate AI text critically risks inaccuracies in scholarly publications, like blindly trusting an inaccurate compass.
Over-reliance on AI without thorough review compromises scholarly rigour. Harnessing AI’s potential as aids to human intellect ensures they complement expertise and discernment rather than replace it.
Promoting ethical disclosure and responsible use of AI-assisted technologies in academic writing
To address concerns of unethical use of AI in academic writing, researchers are actively developing remedies and detection methods. Additionally, academic journals are implementing processes to ensure transparent AI usage disclosure . While countermeasures are essential to mitigate unethical practices, the responsibility ultimately lies with authors to uphold the highest standards of integrity and ethics in their use of AI tools. By remaining diligent and transparent about AI implementation, researchers can contribute to a more responsible and trustworthy academic landscape, ensuring that AI’s potential benefits are harnessed while minimising its risks.
However, as AI tools become increasingly integrated into the writing process, researchers must navigate new ethical considerations, and the issue of their usage and the required level of disclosure becomes increasingly urgent. For example, many of us have relied on the use of basic AI tools, such as grammar and spelling correction features commonly found in modern word-processing software, for years without disclosure. These commonplace AI utilities are integral to enhancing writing quality. While tools like ChatGPT can be used for similar purposes it is important to acknowledge that they are far more advanced and may also impact the substantive content of the article during the editing process. As a result, their use could warrant disclosure to ensure transparency and ethical reporting . Similarly, when seeking feedback or collaborating on manuscripts, it may become necessary to inquire whether collaborators have used such tools to edit or generate any text, ensuring proper disclosure and compliance with reporting policies.
Less-discussed is the “grey area” of AI-assisted review, where permission to use AI tools on others’ work could be debated. Risks emerge when unpublished content like manuscripts or grants are uploaded for AI processing. This could expose sensitive data and even incorporate unfit findings into training datasets, perpetuating misinformation. Ensuring confidentiality and fair assessment calls for avoiding use of AI tools on unpublished content undergoing peer-review. Instead, rigorous adherence to traditional peer review is recommended. The US National Institutes of Health has issued similar guidance, barring reviewers from using AI tech for assessments . Such policies may serve to mitigate the risk of the echo chamber effect, where biases in AI-assistance tools might go unnoticed if reviewers rely on the same tools for evaluation (Table 1).
However, this is not to say that AI-assisted technologies have no utility in providing feedback on others’ work. For example, a researcher may write a rough draft of their thoughts and then “polish” this using a large-language model for clarity and structure. Alternatively, a reviewer could provide a straightforward evaluation and request the tone to be transformed into one of constructive criticism before sending it back to the recipient. Here, one still needs to bear in mind confidentially and, as such, refrain from including identifiable or sensitive data. It is helpful to consider whether you would be displeased by someone revealing similar information about you to a stranger or if sharing such data could in any way harm the individuals being evaluated - now or in the future. If the answer is “yes” to either, then it is best not to share that information with AI tools.
Authors, reviewers, and journal editors share the responsibility of maintaining the highest standards of ethical conduct in the peer review process. As the scientific community embraces the integration of AI technologies, it is essential to balance the advantages of efficiency and accuracy with the ethical considerations surrounding confidentiality and privacy .
Transparency in use of AI-assistance tools
To address these challenges, the International Committee of Medical Journal Editors (ICMJE), which is endorsed by Springer Nature—publisher of BMC Medicine—mandates comprehensive disclosure of AI technology usage in all submitted manuscripts (Table 2). It is import to underscore that AI technologies cannot be acknowledged or credited as authors of articles [7, 8], as they do not fulfil the fundamental ICMJE authorship criteria, which encompass the responsibilities of taking ownership for the published work, declaring potential competing interests, and participating in copyright and licensing agreements. Instead, the onus falls on the human authors to fully assume responsibility for ensuring the accuracy, authenticity, and integrity of all AI-generated content. Citing AI-assisted technologies as primary sources should also be avoided. Adherence to these guidelines will preserve scholarly integrity and scientific rigour in academic publishing.
The updated ICMJE recommendations  thoughtfully address AI’s potential consequences in scholarly publishing. Adoption must proceed cautiously, considering limitations in generating misinformation. Transparency, accountability, and ethical use should guide the development and integration of AI-assisted technologies. While AI can assist in various processes, human creativity, curiosity, and ingenuity remain distinctive and invaluable qualities in science and scholarship that will serve as the bedrock of these disciplines for years to come.
Availability of data and materials
International Committee of Medical Journal Editors
Stokel-Walker C, Van Noorden R. What ChatGPT and generative AI mean for science. Nature. 2023;614(7947):214–6.
Brainard J. Journals take up arms against AI-written text. Science. 2023;379(6634):740–1.
Donker T. The dangers of using large language models for peer review. Lancet Infect Dis. 2023;23(7):781.
Peng Y, Rousseau JF, Shortliffe EH, Weng C. AI-generated text may have a role in evidence-based medicine. Nat Med. 2023;29(7):1593–4.
Liebrenz M, Schleifer R, Buadze A, Bhugra D, Smith A. Generating scholarly content with ChatGPT: ethical challenges for medical publishing. Lancet Digit Health. 2023;5(3):e105–6.
De Angelis L, Baglivo F, Arzilli G, Privitera GP, Ferragina P, Tozzi AE, et al. ChatGPT and the rise of large language models: the new AI-driven infodemic threat in public health. Front Public Health. 2023;11:1166120.
International Committee of Medical Journal Editors. Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly work in Medical Journals (Updated May 2023). 2023. Available from: https://www.icmje.org/icmje-recommendations.pdf. Cited 25 July 2023.
Hosseini M, Resnik DB, Holmes K. The ethics of disclosing the use of artificial intelligence tools in writing scholarly manuscripts. Res Ethics. 2023. https://doi.org/10.1177/17470161231180449.
Meyer JG, Urbanowicz RJ, Martin PCN, O’Connor K, Li R, Peng PC, et al. ChatGPT and large language models in academia: opportunities and challenges. BioData Min. 2023;16(1):20.
Hosseini M, Horbach SPJM. Fighting reviewer fatigue or amplifying bias? Considerations and recommendations for use of ChatGPT and other large language models in scholarly peer review. Res Integr Peer Rev. 2023;8(1):4.
Declaration of generative AI and AI-assisted technologies in the writing process
ChatGPT July 20 Version (GPT3.5) was used during the preparation of this work, specifically to generate similes in the “Recognising the current limitations of large language models” section of this editorial, produce the text outlining a hypothetical example of naïve usage of large language models contained in Table 1, and refine an initial title idea. After using this tool, the author reviewed and edited the text as needed and takes full responsibility for the content of this editorial.
Ethics approval and consent to participate
Consent for publication
SB is currently an editorial board member of BMC Medicine and associate editor at the Journal of Epidemiology and Community Health.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Bell, S. The write algorithm: promoting responsible artificial intelligence usage and accountability in academic writing. BMC Med 21, 334 (2023). https://doi.org/10.1186/s12916-023-03039-7