Several papers that have recently been published in peer-reviewed journals display obvious signs of having been written by the AI tool ChatGPT. This has sparked a heated online debate about the transparency of research communication and academic integrity in cases where AI is used in the academic writing process. In this blog article, Kim Tung Dao discusses the ethical implications of using AI for academic writing and ponders the future impact of AI in academic research, urging for a balance between the efficiency of AI tools and research integrity.
Used for everything from streamlining everyday tasks to revolutionizing industries, artificial intelligence (AI) has come to profoundly affect our lives in the past few decades. The emergence of new forms of AI in recent years has led to a heated debate in academia about whether students should be allowed to use AI tools — usually large language models (LLMs) such as ChatGPT — in their writing. And if they are permitted, a related question is to what extent they should be used, especially in higher education.
A new issue related to the rise of LLMs is now rearing its head within the realm of scientific research: the publication of LLM-generated content in peer-reviewed journals. This worrying trend reflects not only the rapid advancements in LLMs’ ability to replicate human work but also gives rise to discussions on the ethics of research (communication) and research integrity.
More and more researchers are attempting to leverage generative AI such as ChatGPT to act as a highly productive research assistant. It is very tempting to have an LLM compose content for you, as these AI-generated pieces often exhibit sophisticated language, conduct statistical analyses seamlessly, and even discuss new research findings expertly. The line between human- and machine-generated content is blurring. In addition, these LLMs work tirelessly and quickly, which can be considered highly beneficial for human scholars.
However, beneath the surface of effectiveness and efficiency lies a complex labyrinth of ethical concerns and potential repercussions for the integrity of scientific research. Publishing academic research in journals remains the most popular way for many researchers to disseminate their findings, communicate with their peers, and contribute to scientific knowledge production. Peer reviewing ensures that research findings and truth claims are meticulously evaluated by experts in the field to sustain quality and credibility in the formulation of academic theories and policy recommendations. Hence, when papers with AI-generated content are published in peer-reviewed journals, readers can’t help but question the integrity of the entire scientific publishing process.
There is a big difference between receiving assistance from generative AI and allowing it to generate entire or significant parts of research texts without appropriate supervision and monitoring. These can entail smaller tasks such as proofreading AI-generated content before its distribution/publication but can also play a much more critical role in ensuring the originality and significance of AI-enhanced research. This is why this article seeks to reflect on the abuse of AI in the writing of academic texts by researchers and provides commentary on the insufficiency of the current peer-review system. I also try to initiate a thoughtful discussion on the implications of AI for the future of research.
Falling through the cracks
The latest volume of Elsevier’s Surfaces and Interfaces journal recently caught the attention of researchers on X (Twitter), as one of its papers has evidently been written by ChatGPT. The first line of the paper states: “Certainly, here is a possible introduction for your topic: […].” Any ChatGPT user knows that this is the typical reply generated by the LLM when it responds to a prompt. Without any expertise in AI or other related fields, a common ChatGPT user with normal common sense can therefore tell that this sentence and at least the following paragraph, if not many others, has been generated by ChatGPT.
But this paper is certainly not the only one in this new line of LLM-generated publications. ChatGPT prompt replies have been found in other papers published in different peer-reviewed journals and are not limited to any specific fields of science. For example, a case report published in Radiology Case Reports (another Elsevier journal) includes a whole ChatGPT prompt reply stating “I’m very sorry, but I don’t have access to real-time information or patient-specific data, as I am an AI language model. I can provide general information about […], but for specific cases, it is essential to consult with a medical professional […].”
Hallucinating information
What is more worrisome is the quality, integrity, and credibility of scientific research conducted by these LLMs, as ChatGPT has the tendency to hallucinate information and draws on seemingly non-existent citations and references to support the texts it generates. For example, in a forum discussion where contributors talked about detecting AI-generated content in academic publications, one contributor pointed out that they could not find the references cited in a paper titled “Automatic Detection of Coagulation of Blood in Brain Using Deep Learning Approach”. Several other cases are mentioned in the discussion thread.
Besides likely contributing to the publication of false or unevidenced information, the use of LLMs in the writing up of scientific research also highlights the failure of peer reviewers to catch or question these practices, showing either their carelessness or their irresponsibility. The peer-review system has long served as the gatekeeper of scholarly knowledge, aiming to uphold high standards of quality, integrity, and credibility that are part and parcel of academic research and publishing. But with obvious evidence of LLM-generated content being included in papers published in peer-reviewed journals, it might be time to start questioning the transparency and accountability inherent in the peer-review process. When a peer-review publication starts with a ChatGPT’s typical prologue, it’s reasonable to wonder how such article was reviewed.
A call for responsible use
AI is not all bad. Clearly, it can be a powerful assistant to researchers in the research process, used for anything ranging from brainstorming, developing research strategies, coding, analyzing empirical results, and language editing to acting as a competently critical reviewer to provide useful and helpful feedback for excellent improvement. But to work with this powerful assistant, researchers still need to have a solid knowledge of the research topic, make significant decisions on the research strategy, and, most importantly, ensure that the research is an original contribution to the literature and can be applied. Relying heavily on AI to finish a research project without understanding the foundation and the essence of the research is plainly ethical contamination and fraudulent behavior.
AI is not a scientific researcher — and might never be
Beyond the immediate finger-pointing at the peer-reviewed system and research practices, the increasing influence of AI in research outputs carries broader implications for the role and integrity of human researchers, the nature of scientific discovery, and the social perception of AI. Even if the potential for deception and manipulation is ignored, AI-generated research outputs might still lack genuine insights, critical analysis, and might fail to take into account ethical considerations without human guidance. Moreover, in order for research outputs to be meaningful for human life and society, they need to be validated by human researchers.
We don’t necessarily need to fear AI; we do need to fear the improper use of AI, and we need to play an active role in preventing this from happening. Thus, instead of fearing being replaced by AI, human researchers should start acknowledging its abilities and using it to shape our projects. Let’s board this technological advancement ship to escalate our research efficiency and accelerate the speed of scientific discovery. But let us remain cautious. We are responsible for ensuring that AI contributes to instead of compromises scientific knowledge production.
Writing this post with the help of ChatGPT 3.5 (which I used to improve my language), I can’t help but recall the question I was asked when receiving my doctoral degree: “Do you promise to continue to perform your duties according to the principles of academic integrity: honestly and with care; critically and transparently; and independently and impartially?”
I promise.
Opinions expressed in Bliss posts reflect solely the views of the author of the post in question.
Are you looking for more content about Global Development and Social Justice? Subscribe to Bliss, the official blog of the International Institute of Social Studies, and stay updated about interesting topics our researchers are working on.
[newsletter]