Editor’s Note: Today’s post is by Marco Marabelli, Robert M. Davison, and Giovanni Gatti. Marco is Professor of Information Systems and Process Management at Bentley University. Robert is Professor of Information Systems at City University of Hong Kong. Giovanni is a PhD candidate in Management and Innovation at Università Cattolica del Sacro Cuore. Reviewer credit to Chef Alice Meadows.

Generative AI, Academic Research, and Publishing

The role of Generative AI (GenAI) has lately been a hot topic for The Scholarly Kitchen’s readers, particularly for its transformative power in the publishing industry, with guest posts discussing the balance between leveraging emerging technologies and producing human-centric research, for instance. But let’s step back for a second. It all began in 2017, with Vaswani and colleagues’ seminal article, “Attention is All You Need”, which disrupted the AI space. They introduced new ideas to explain how AI processes language, with a focus on novel AI systems (i.e., GenAI) that were able to receive inputs (prompts) from users that were used to generate seemingly coherent outputs (text, images, music). Individuals, organizations, and society have since been heavily impacted and, as generally happens with disruptive technologies (cf. television, mobile phones, smartphones, the internet), the consequences, both positive and negative, have exceeded the wildest predictions.

Among many aspects of GenAI concerning academics, one in particular sticks out: the extent to which GenAI should (or should not) be embedded in various steps of the research process, and associated ethical concerns. Initially (in 2021/2022, when the first popular GenAI system, i.e., ChatGPT-3.5, developed by OpenAI, was released to the general public), GenAI was found helpful to “automatically” copy-edit manuscripts. However, more recent developments illustrate how GenAI can now perform a literature review (even if with caveats), analyze data, formulate questions and hypotheses, and even write a whole paper from scratch. Recommendations for further applications are proliferating.

In our opinion, the use of GenAI for research is highly problematic for many reasons. First, GenAI hallucinates, i.e., it creates outputs that may seem plausible yet are manifestly and materially inaccurate. Second, GenAI is trained with data that is inherently biased, notably internet-derived sources. Thirdly, reliance on GenAI may lead to what we call scholarly deskilling, or the progressive inability to undertake such tasks as writing a research paper without GenAI’s help. GenAI may also influence the essential intellectual contributions (ideas, creativity) central to a paper that we might otherwise imagine to be the sole preserve of human beings. Fourthly, in our own experience, we realized that many scholars (especially junior scholars) currently using GenAI are either unaware of or complacent about these issues, and, more importantly, not all are aware of how these systems actually operate. One consequence of these problems is a torrent of GenAI-themed and influenced research that is minimally valuable to society.

We are aware that there are now several GenAI tools available to assist researchers in essentially every step of the research process and that some scholars support their use, for instance, to aid theorizing. We are also aware that GenAI use for research is an emerging topic and, at the time of writing, not all scholars are familiar with either the opportunities or the challenges associated with the use of such tools. We therefore decided to organize a panel, in October 2025, at Bentley University in the US, where we invited publishers (Elsevier, Emerald, Gailot Press, MIT Press, Nature/Springer, Wiley) and editors in chief (representing the Information Systems Journal, Journal of the AIS, Journal of Data Scienceand ICES Journal of Marine Science) to discuss the pros and cons of using GenAI, along with ethical and policy implications. We aimed to discuss current ethical concerns associated with using GenAI in research and wanted to realize the extent of the alignment (or misalignment) between editors and publishers.

Main Takeaways

The first takeaway from the panel discussion was that the topic of using (or not using, or using to what extent) GenAI for research is a contested one, with the participating publishers and editors exhibiting different, even diametrically opposed, perspectives. Given the increased “digitalization” of journals and books, academic publishers’ business models are very much centered on “quantity” (as long as quality can be maintained) as a reflection of open access, as traditional print subscription models are fading. In contrast, editors see quantity as a problem. As already alluded to above, crafting manuscripts with GenAI makes the writing process faster, especially if tasks such as literature review and data analysis are partially (or wholly) outsourced to automated systems. Publishers and editors participating in the panel seem to be on the same page concerning the urgency to focus on full disclosure of GenAI use (by authors) and on the review process that may become problematic if outsourced (even in part) to GenAI because of issues associated with intellectual property and privacy of study participants. However, it is likely that, in the near future, the volume of paper submissions will increase to the point where there are inadequate human resources to deal with the flood in a timely fashion.

The second takeaway was that GenAI is here to stay. While the panel participants had different positions on the extent to which GenAI should be used, all agreed (some fervently, some reluctantly) that it is unreasonable to think of banning GenAI use in full. It is important to study sociotechnical implications of emerging technologies mingling with various steps of the research process. We believe that various research disciplines (and their associations) should focus on the ethical implications of using GenAI for research. These include academic misconduct, i.e., scholars deviating from publishers’ and editors’ policies (and the associated issue of the lack of effective GenAI detection systems); and considerations concerning the extent to which automated systems such as GenAI, which are trained with biased datasets, should participate in the literature review and data analysis components of the research process.

Regarding academic misconduct: 1) How can scholars comprehensively train new generations of PhD students and junior faculty to conduct research ethically, while living in a “publish or perish” environment? The temptation to cut corners by outsourcing some research tasks to GenAI is strong, especially given the absence of reliable detection systems. 2) What could or should be the consequences for academic misconduct involving GenAI use that goes beyond stated policies? Would this misconduct be treated as similar to plagiarism or as something even more serious? The first question will involve rethinking mentoring programs in schools and at major academic conferences (for instance PhD and junior faculty consortia), to raise awareness of the importance of conducting research ethically. The second question is more difficult to answer because, as we noted earlier, it is difficult to establish, with a suitable degree of certainty, whether someone has used GenAI for research, and, if so, to what extent. This issue is particularly important in reference to the review process, where publishers and editors seemed to agree that outsourcing a review to GenAI is inappropriate.

Regarding GenAI’s inherent biases: how (if it is at all possible) can researchers vet GenAI outcomes and ensure that they do not embed implicit biases? Here, the problem is that researchers themselves are biased. It may be almost impossible for a white male researcher to spot a bias in outputs generated by a GenAI system built on a Large Language Model (LLM), trained predominantly on white, male-based data (e.g., Wikipedia). This problem goes beyond the idea of simply keeping a “human in the loop” in the review process, as we argued in a recent paper. It also begs the question of whether it makes any sense to have a human vetting a tool where both parties may share the same underlying biases, especially when the system itself cannot be held accountable due to its lack of human agency.

The third takeaway reflects how both publishers and editors acknowledging the constantly evolving nature of GenAI technologies, and the subsequent need to update GenAI policies at the publisher and editor level. For instance, what if a paper is desk rejected by an editor because the authors’ declaration of GenAI use is not compliant with “current” policies, but then these policies change and become more indulgent the next month, such that had the paper been submitted a month later it would not have been desk rejected? Would this constitute fair practice for a journal (or unfair treatment for the submitting author)? Related to this is an issue associated with the inconsistent policies that prevail across publishers, which was discussed by the panel: would journals and publishers with “stricter” GenAI policies be viewed as being more legitimate or credible? Would they receive fewer submissions?

Overall, it would be meaningful for scholars to be able to conduct cross-disciplinary studies on journals and publishers and perform statistical analyses on the extent to which GenAI policies affect the two year Impact Factor, reject ratio, and other metrics. In addition (medium-term target), it would be meaningful to discover, statistically, correlations between acceptances and rejections and the two year impact of a published paper (i.e., citations affecting a journal’s Impact Factor) vs. the authors’ declaration of GenAI use. Will GenAI use make a difference in terms of paper acceptance and impact, and if so, will GenAI affect acceptance and impact positively or negatively? It is important to note that it is possible that the GenAI policies of specific editors and journals may not be aligned with those of their publishers. From the panel and our own observations, if editor and publisher policies diverge, specifically regarding rules on GenAI use, the editor tends to apply stricter standards. This is understandable, since editors have to shoulder the burden of increased submission volumes and, in addition, have fewer financial incentives than the publisher, which focuses on publishing as many (high-quality) papers as they can.

Lastly, an important takeaway that emerged from the panel is that it remains unclear how far GenAI will affect future scholarship, particularly regarding an individual’s ability to retain their research skills. Several studies have now demonstrated that reliance on certain technologies can lead to dependency and scholarly deskilling. For instance, relying on GPS systems to drive a car has decreased our sense of spatial orientation. More broadly, as individuals increasingly depend on automated support rather than actively exercising and refining their own abilities, forms of cognitive disengagement emerge, driven by the convenience afforded by algorithmic or AI systems. In the context of GenAI, emerging evidence suggests that when problem-solving activities are delegated to these systems, individuals tend to perceive tasks as being less cognitively demanding. This delegation is also associated with reduced neural connectivity and lower levels of engagement compared to non-users.

Here academics should reflect on whether it is acceptable to welcome GenAI into the research process, given that it is likely that our research skills will deteriorate in direct consequence. What happens if a technology breakdown occurs, i.e., when GenAI stops working, even temporarily, and someone lacks the capability (due to reliance on GenAI) either to put together a PowerPoint presentation or to address the final requests from a reviewer before submitting a paper? What is more, both our own knowledge of published works studying the GenAI phenomenon in academia, along with theoutcomes from the panel, seem to lead to the shared view that GenAI should not be used in the review process. If the review process is, even in part, delegated to GenAI, how can PhD students and junior scholars learn how to write papers when they aren’t engaging deeply in the review process? Besides the potential deskilling (or “no-skilling”, for PhD students and junior faculty), let’s remember that uploading papers or even reviews of papers in public, or “external GenAI systems such as Chat GPT (versus Copilot for enterprises, which could be considered “internal”)      can represent an intellectual property violation (depending on the material’s copyright status and license terms). In some cases, it also raises privacy issues, for instance, if the paper being reviewed contains direct quotes from study participants.

Overall Reflections

To summarize, many aspects of GenAI use for research remain very much “in flux”, either because publishers and editors hold different perspectives, or because it is not yet clear how GenAI technology will evolve in the short and medium term. We believe that scholars versed in studying sociotechnical systems are probably among the best-positioned academics to explore the implications of GenAI for research, because it involves people, processes, and organizations (where various practitioner actors read, or should read, our research outputs), all of which are deeply intertwined. While restrictive regulation of GenAI is highly unlikely, individual scholars, institutions, editors, journals, and publishers are all likely to develop their own standards for what is appropriate. At the same time, it is plausible that industry-wide norms or formal standards may emerge, potentially driven by international standard-setting bodies such as ISO and IEC, as reflected in ongoing initiatives (e.g., ISO/IEC AWI 25590). Such standards may be framed as policies or principles, depending on the actor. The extent to which these policies and principles converge and diverge will surely influence the extent to which GenAI has an impact on the research process. Even in our own community (information systems), we are aware of journal editors with diametrically opposed principles, some encouraging all forms of GenAI use, others forbidding it for all intellectual tasks in equal measure.

A more comprehensive report of the October 8, 2025, panel was published by Communications of the Association for Information Systems and is available here.

Authors’ note: The full reference of the article this post draws upon is: Marabelli, M., Davison, R. M., Gatti, G., Srivastava, A., & Tarafdar, M. (In press). Use of Generative AI in Scholarly Research: Challenges and Opportunities. Communications of the Association for Information Systems, 58, Accessible at https://aisel.aisnet.org/cais/vol58/iss1/36