Introduction
During the DESIGN 2026 Conference on 19 May 2026 in Cavtat, Croatia, the Design Science Journal hosted an open panel on AI and Publishing Scientific Papers: A Discussion with Design Journal Editors. The panel session explored how artificial intelligence (AI) is transforming scientific publishing. Panos Papalambros, Editor-in-Chief and Anja Maier, Co-Editor-in-Chief of the Design Science Journal, moderated the panel. The panelists were Editors-in-Chief of eight design journals:
-
• Jonathan Cagan, Co-Editor-in-Chief, Design Science Journal.
-
• Gaetano Cascini, Editor-in-Chief, International Journal of Design Creativity and Innovation.
-
• Amaresh Chakrabarti, Editor-in-Chief, Artificial Intelligence in Engineering Design and Manufacturing (remote).
-
• Lin-Lin Chen, Editor-in-Chief, International Journal of Design (remote).
-
• John Gero, Editor-in-Chief Emeritus, Design Science Journal.
-
• Roger Jiao, Editor-in-Chief, Journal of Engineering Design (remote).
-
• Peter Lloyd, Editor-in-Chief, Designing.
-
• Jianxi Luo, Editor-in-Chief, Research in Engineering Design.
-
• Carolyn Seepersad, Chief Technical Editor, ASME Journal of Mechanical Design (remote).
The panelists shared their initial thoughts and engaged in discussion with the audience. The session encouraged an open dialogue among researchers and editors on current and future implications and impact of AI on publishing scientific papers, particularly in design. Topics included conduct and reporting of research, paper composition, reviewing and referencing, the economic and social cost of knowledge generation and production, skills development, AI in research and supervision more widely, the role of humans and implications for humanity, ethical and policy questions.
Following the conference panel session, panelists were asked to provide a 500-word perspective on the panel topics, which the moderators assembled into the present editorial. The panelists were given no instructions other than the word limit and a limit on citations in their piece. Their thoughts are included below in alphabetical order.
Jonathan Cagan
To start, I am not opposed to research on AI or Machine Learning (ML) – how to make engineering and particularly the engineering design process more efficient and effective. I am supportive of research on how AI can perform better than humans even in creative situations, or just how AI can help human designers. However, research with and on AI is different from using AI to write or even significantly augment publications. It is different from using AI to make conclusions, even when it seems straightforward to do so.
For example, we have seen a significant growth in “review papers.” Such papers were thoughtful, requiring effort to accumulate and assimilate their lessons and derive insights that would help others in the field. However, given the sheer volume of such papers submitted today, the majority must “use” AI in some way in assimilating, assessing and writing the review.
I recently took the top 10 papers from the DESIGN 2026 Conference Proceedings with the highest review scores, uploaded them to a closed version of ChatGPT (closed in that it was local and not used to train the Large Language Models (LLM)), and asked it simply for common insights that capture all 10 papers (randomly selected in practice). This is what it told me:
There is a shift in design research: from designing isolated products toward designing evidence-based interventions in socio-technical systems. Whether the subject is AM, healthcare, infrastructure, sustainability, patents, or education, the common finding is that better design requires earlier modeling, richer data, stakeholder/system awareness, and humility about what tools – especially AI – can currently do.
This could easily be the basis for a review paper. I have even looked at the research questions from a recent submission of a review paper, entered those questions and asked an LLM to look for papers in the field and answer them. The result was more insightful than the submitted paper, which again did not acknowledge use of an LLM but was so shallow it clearly had. So, the bar for a review paper to be meaningful has been raised quite a bit.
At a minimum, and more generally, authors need to be honest and appropriately say how exactly they are using LLMs. This is an ethical requirement that all authors must follow. But even further, if an LLM is used to heavily write a paper, then what role do we as researchers play?
Looking at the future world of publishing, I ask, but do not answer, several questions we must grapple with as a community:
-
• With authors’ use of LLMs in writing, are there legal implications for journals, reviewers and authors?
-
• Do we accept papers that have used LLMs as is? Should we require joint authorship with the LLM?
-
• Do we interview authors before accepting a paper to confirm that they actually wrote and understand it?
-
• Do we require demonstrable impact of the research, that is instantiation?
-
• Are there alternative paths for dissemination? Will papers, as a means to communicate research, become irrelevant?
-
• At the same time, how do we as a research community and as those responsible for our journals (editors, reviewers and authors alike) celebrate these new capabilities as we look to the future body of research dissemination for our field?
Gaetano Cascini
The discussion on AI and publishing often starts with anecdotes about authors using AI, reviewers relying on AI or editors receiving AI-assisted manuscripts and reviews. These examples are illustrative but can obscure the key phenomenon: AI is lowering the cost of producing papers. It supports drafting, language polishing, figure preparation and compliance with journal requirements, improving accessibility for authors across languages or disciplinary boundaries.
A different question is whether AI is lowering the cost of producing new knowledge. It may do so by organizing literature for state-of-the-art analyses, elaborating data, recognizing patterns or comparing large information sources, provided that researchers can verify outputs. Yet this advantage has not been demonstrated with the same clarity as AI’s effect on manuscript production.
The likely near-term consequence is an imbalance: for a given amount of effort invested in knowledge creation, more manuscripts can be produced. This is beneficial when it improves communication, but problematic when it interacts with academic careers, university rankings and funding systems that reward publication counts and measurable outputs over the value of knowledge produced.
Without changes to publication processes, AI may amplify publish-or-perish dynamics. Editorial experience suggests increasing submission volumes and growing pressure on editors and reviewers. AI tools may help editorial workflows, but they cannot remove the asymmetry: thousands of AI-assisted authors can generate more manuscripts than a much smaller number of AI-assisted editors and reviewers can responsibly assess. The issue cannot be addressed only through declarations, ethical statements or detection tools, which often prove unreliable. It requires changing the rules of the game.
One possible direction is to give renewed value to scientific interaction. If AI increases the volume of publishable-looking manuscripts, human dialogue becomes more important. Conferences, editorial forums and journal processes could create more space for discussion, critique, replication reports and visible scholarly credit for reviewing. Conferences might select papers for longer presentation and discussion, with clearer pathways from substantial debate to journal publication. The aim is not to make conferences more exclusive, but to examine contributions more deeply and keep human dialogue central to knowledge validation.
A second direction is to strengthen replication, extension and explicit connections among published works. Current publishing cultures often privilege isolated novelty claims, while giving less recognition to checking, reproducing, comparing or building cumulatively on previous results. In an AI-rich environment, replication becomes necessary for distinguishing fluent scientific text from reliable scientific knowledge.
A third direction concerns process transparency. Manuscripts should make it easier to understand how knowledge was produced: what data, methods, assumptions, prompts, models, code or analytical decisions shaped the result, where relevant. This is especially important in design research, where inquiry is situated, constructive, experimental and methodologically diverse. We are already moving in this direction, but journals could push further to enable faster and more reliable human assessment.
In conclusion, AI should not be treated simply as a threat to publishing, nor only as a productivity tool. It is also a reason to redesign scholarly publishing itself: its incentives, reviewing practices, conference formats and mechanisms for cumulative knowledge building.
Amaresh Chakrabarti
There are two major questions on the role of AI in publishing:
-
• How can AI be used to support an author in preparing a manuscript for research or publication?
-
• How can AI be used to support a reviewer or editor in evaluating a manuscript for publication?
We clarify below some major, current stipulations on the above and make some general observations. The work is gleaned from multiple sources, for example COPE, APA and AI ethics guidelines of some major journals, such as
-
• https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools
-
• https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals
-
• https://journals.asianresassoc.org/index.php/ajir/ai-policy
-
• https://www.apa.org/pubs/journals/resources/publishing-tips/policy-generative-ai
AI in supporting research and preparation of a manuscript
The following are fairly consistent across policies of various journals and organizations:
-
• AI Cannot be an Author: The rationale is that AI lacks legal standing and cannot take intellectual responsibility for the correctness, originality or potential biases of the work.
-
• Only Authors are Accountable: The author retains sole responsibility for the content; since AI can “hallucinate,” authors must manually verify all information used from AI.
-
• Mandatory Transparency and Disclosure: Most journals need authors to declare how and where AI is used. Simple grammar or spell-checks are usually excluded from this.
-
• Restrictions on Data and Images: AI can be used to analyze data in the methodology stage, but this must be documented in detail for reproducibility; however, use of AI to create or manipulate primary data, charts and evidential images is prohibited.
AI in supporting reviewing of manuscripts
The following are some major points from policies of various journals:
-
• Peer Review Prohibitions: Peer reviewers are prohibited from uploading submitted, confidential manuscripts into AI tools to assist in the peer review process. Cambridge University Press (CUP), for instance, insists on four aspects in peer reviews: confidentiality (of work and authors), transparency (reviewers must declare their use of AI), accountability (editors are responsible for all decisions and explanations about the suitability of submissions) and fairness (reviewers/editors are responsible for managing use of information from AI to avoid possible cultural, disciplinary, linguistic and other bias).
-
• Peer Review Provisions: What editors and reviewers can do safely is to: use AI to organize their thoughts and prioritize feedback; use AI to improve review clarity and tone or translate the review; ask AI general questions about research methods or statistical approaches without referencing manuscript specifics.
General observations
Some general observations and open issues are the following:
-
• The onus of execution of the policies rests on authors, reviewers and editors, with very few reliable processes, tools, checks and balances available to them to ensure this.
-
• Given the increasing pressure on all stakeholders to “publish or perish,” some are likely to stray, intentionally or otherwise, which is often difficult to identify or prevent.
-
• As AI itself continues to evolve, even if such tools are created, their relevance would be relatively limited.
In conclusion, we need more such discussion in research fora and across research fora, journals and policy organizations, to evolve robust policies and their operationalization.
Lin-Lin Chen
Journal editors face a key challenge in the age of AI: how to handle the huge growth in submissions. Like at many other journals, I am observing a threefold increase in the number of submissions to the International Journal of Design (IJDesign). On the other hand, sadly, I have not seen a significant change in terms of the number of good quality submissions. During the last year, I encountered one submission that had clearly fake references and one submission where it is likely that an AI tool was used to identify “suitable” references previously published in IJDesign.
This huge growth in submissions puts enormous pressure on the peer review process. All editors know that a high-quality journal relies on the voluntary contribution of peer reviewers who are experts in specialized research areas. The time and capacity of these experts is highly limited and can never match the speed of such growth in submissions. Since IJDesign does not plan to use AI tools in pre-screening submissions, the role of initial review has become ever more important in screening out submissions that do not fit the scope and focus of the journal or meet reasonable academic quality standards, so as not to overburden the peer review process.
When AI tools are used well, I believe that they can help elevate the quality of manuscripts. But how can AI be used in an ethical and responsible way? I am glad to see that the International Association of Scientific, Technical & Medical Publishers (STM) published guidelines for the use of generative AI. Their “9 Recommended Classifications of AI Activities” offer more nuanced distinctions between different ways that AI is used. On the other hand, most of us are both authors and peer reviewers. We should be able to reason from both perspectives and develop consensus on the fair use of AI tools. As authors, we should be explicit and clear about how and where AI tools have been used. We might also explicitly identify sections where AI tools have NOT been used. Such information will be helpful to peer reviewers when evaluating manuscripts. As reviewers, we should be aware that uploading an unpublished manuscript to AI tools for analysis is a violation of intellectual property.
For authors who are not native English speakers, AI tools are helpful in producing well-translated manuscripts. This alleviates the language barriers faced by many authors, enabling them to share their work with a wider academic audience. However, what determines a well-written manuscript is not just the language but, even more importantly, the actual content of the work. With the availability of AI translation tools, I have seen many manuscripts that are well-translated in terms of English, but whose contents remain disconnected from the up-to-date design research literature. I am hoping that such issues will gradually disappear with much better AI tools and awareness in the future.
John Gero
Academic journals’ existence is founded on two principles: to provide a vehicle to make new knowledge publicly available and to attribute that knowledge to its intellectual authors. To ensure the knowledge is “new” they empower external peer reviewers to attest to its originality. The use of AI in publishing research papers in journals should be assessed against these two principles.
Let us start with the new knowledge contained in a paper. Research results are a consequence of research questions or hypotheses developed by the authors. Various research methods are employed to produce source data that is used to examine the research questions or to test the hypotheses. These methods need not be decided on or deployed by those who developed the research questions or hypotheses. The analyses of the source data need not be carried out by those who produced the source data. This separation of activities often results in multi-authored papers, although not all contributors are recognized as co-authors depending on their level of contribution. Different disciplines use different criteria for inclusion as authors. Design research papers generally have few authors, while neuroscience papers have a multitude of authors. In addition to these activities, there are others where people are involved in the development and production of a paper. These include, for example, coders, graphic artists and copy editors, none of whom are counted as co-authors in most disciplines.
Viewed in this articulated way, papers require many contributions, and we can examine where AI may be deployed under human oversight and where it might be problematic. Copy editing, that is polishing the language style, can be carried out by an AI without any compromise to the endeavor. It replaces an already unattributed human activity. Similarly, the production of charts and diagrams can be carried out by an AI without any compromise to the endeavor. Coding, whether cognitive coding of transcripts or coding in the form of computer programming, can be carried out by an AI without compromising the endeavor. These areas of contribution by an AI are not particularly controversial. Let us move on to grayer areas.
What is the difference between sending a research assistant to locate appropriate references and asking an AI? I suggest there is no fundamental difference. Both would require human supervision. Similarly, the same would apply in having an AI decide which statistical method is appropriate for the goal and the available data. Again, this would require human supervision. What about designing an experiment in response to research questions or hypotheses? Research plans are published to allow the reader both to assess their viability and to reuse them. This is an even grayer area for many people. From the perspective of the paper and its reviewers, it does not matter who produced the research plan; it is the quality of the plan that should matter. Where this matters is in the responsibility for the plan, which should remain with the author.
This leaves two further areas: motivation and research questions/hypotheses. The motivation for the research being reported comes from the researchers. However, when prompted, an AI can readily provide 10 research gaps in most design research fields, which could serve as a basis for a researcher’s motivation. Similarly, for the AI and research questions. What then is the role of the researcher and the AI in terms of the publication?
If a paper used an AI in all its research phases and was passed successfully by multiple peer reviewers, implying the knowledge was “new,” then it should not matter that an AI was used. This leaves the issue of authorship. Authorship can be maintained through transparency and the author’s integrity in stating the level of usage of AI, in the same way that multi-authored papers indicate the relative contributions of the various authors. In the extreme case, the author becomes part symphony conductor and part score writer.
Roger Jiao
AI recentralizes design research around problem formulation and contextual validation
The rapid emergence of AI is reshaping the landscape of scientific publishing, particularly in design (e.g., Chen, P., et al., 2026. An LLM-based multi-agent system to assist early-stage product design and evaluation. Journal of Engineering Design 37 (3), 945–980.). Its impact spans the full lifecycle of research, from problem exploration and coding to manuscript composition, reviewing and referencing. While AI technologies bring undeniable efficiency, they also raise a fundamental question: what remains the core intellectual contribution of design research when many technical “how” tasks become increasingly commoditized?
In engineering design research, the answer lies not in methods alone, but more fundamentally in problem formulation. As recent consensus has emphasized (e.g., Jiao, R. 2026. Towards rigorous problem formulation for engineering design research: from motivations to measurable claims via metric-measure-method. Journal of Engineering Design 37 (6), 2145–2171.), problem formulation is not merely a preliminary step but the defining act of research itself, that is shaping constructs, guiding methods and determining what constitutes valid evidence. AI tools, by lowering the barrier to implementing algorithms and generating results, effectively shift the locus of intellectual merit upstream. When coding, modeling and even writing become accessible commodities, the originality and rigor of research increasingly depend on how well a problem is framed, whether it is anchored in a meaningful context, articulated with operational clarity and supported by measurable claims.
This shift reveals a persistent imbalance in many manuscripts, characterized by an overemphasis on methods and results while research framing remains underdeveloped. Studies that appear technically sophisticated may nevertheless lack a clearly defined and meaningful problem context, resulting in contributions that are primarily application- or solution-driven, a typical case of “a hammer looking for nails,” rather than being motivated by a well-articulated research hypothesis. Consequently, the work is often insufficiently justified in terms of its potential to generate new knowledge, advance theory or methodology or provide original intellectual insights. In this sense, AI does not introduce new shortcomings but rather amplifies existing ones. It sharpens the distinction between methodological execution and intellectual contribution, making the latter both more visible and more critical than ever.
At the same time, AI challenges traditional notions of novelty and validation (e.g., Wu, J., et al., 2025. Integrating AIGC with design: dependence, application, and evolution - a systematic literature review. Journal of Engineering Design 36 (5–6), 758–796.). As solution generation becomes easier, novelty can no longer rely solely on demonstrating algorithmic performance or incremental technical improvements. Instead, validation emerges as a central pillar of design research, particularly in relation to context and use (i.e., Type I and Type II validation). The distinction between verification (showing that a method works under controlled conditions) and validation (demonstrating that it matters in real-world settings) becomes increasingly important. Without meaningful validation, research risks producing results that are internally consistent yet externally irrelevant.
This brings renewed attention to a foundational principle in design research, namely that context matters. Design problems do not exist in isolation, but are embedded within industrial, organizational and socio-technical systems that shape their significance and evaluation. Research that abstracts away these contexts may achieve analytical elegance, but risks generating what has been termed an “illusion of rigor,” where results are precise yet disconnected from practice (e.g., dataset-centric design research). In contrast, contextually grounded research, supported by real-world case studies and explicit engineering settings, provides not only validation but also a basis for meaningful problem formulation.
In this evolving landscape, the role of authors, reviewers and editors must also adapt. Authors must treat problem formulation as a first-class research artifact, explicitly linking motivations to measurable claims and validation strategies. Reviewers should evaluate coherence and relevance alongside technical sophistication. Editors, in turn, should reinforce norms that prioritize framing quality, contextual grounding and validation rigor.
Ultimately, AI does not diminish the value of design research; rather clarifies and empowers scholarship. By commoditizing routine aspects of research, AI redirects attention to what has always been central but sometimes overlooked, namely, the ability to define meaningful problems, situate them in real contexts and produce knowledge that matters.
Peter Lloyd
One of my PhD candidates is in the final stages of writing up their thesis and they recently submitted a draft of their first chapter. I read it through and wasn’t sure about some of the sentences. I had that now familiar feeling of beginning to struggle with some of the reasoning. I suspected AI use, but of course, could not be certain. When I was giving my feedback, I asked: “Did you use AI for this chapter?” And they said: “oh yes, sure.” When I pointed out that some of the references did not exist, they said: “Yes, there could be up to 30% that are hallucinated.”
They were quite happy with this. Probably not with hallucinated references, but with the idea that I, as an advisor and editor, would help fix these “AI problems.” I had invested significant work in reading the chapter and it felt like I had been cheated. It also felt like any feedback I gave would be used in a new AI dialogue to generate the next draft of the chapter, which would come back to me very quickly. It felt wrong, but was it wrong?
Guidelines do not really help us with this situation. We can have declarations about the use of AI but we still need to make a judgment about those declarations and the expectations that they might put on reviewers. Transparency does not imply acceptability. Transparency and acceptability sound like they belong together, but they are opposed in some ways. Authors will not be transparent if they think that transparency will lead to desk rejection.
Speaking as an editor, if something arrives in front of me, I tend to assume that the author has acted in good faith, followed guidelines and acted ethically. That they have not created their data, for example or generated their analysis. These are the kinds of research integrity issues that we teach our PhDs. So, when I find submissions that have non-existent references or circular, unfocused reasoning, it saddens me. It feels like we are beginning to lose something as a community.
We should not idealize human-created papers as perfect. There is no perfect paper. All papers have problems. All reviewers will find problems with papers, whether they are human-generated or AI-generated. But the very fact we are having this conversation means we have some awareness of the complex ethics around this. A feeling about what we should be doing and what we should not be doing. But we need to be open and non-judgmental about what we do with AI, to explore, experiment and document. To use new practices to explore ethical issues and reach nuanced judgments. We should not dismiss researchers exploring new techniques, or reviewers exploring different ways to review papers out of hand. Conversations need to be open, but they should also come before paper submission. The point of submission should be a moment of clarity – for editors and for authors.
I sometimes feel a bit guilty when I use AI, as if I am short-circuiting my thinking, trying to replace one of the activities that I actually find most satisfying. But newer generations do not feel this, at least not directly. In some ways that is good: to experiment and try to find new practices of research, writing and design. In spite of the above caveats, my overriding feeling is that design is a subject area that should be at the forefront of experimentation and development in the uses of AI for research, publication and (naturally) design.
Jianxi Luo
Redesigning the Actor-Action-Artifact system of publishing in the AI age
AI is reshaping academic publishing, introducing paradigm-shifting opportunities alongside profound governance challenges. Viewed through the Actor-Action-Artifact (3A) framework of Design Science (Luo, J. 2026. Design Science: Driving Innovation with Unified Principles. Springer.), AI acts as a simultaneous reconfigurator of the actions underlying research and dissemination, the artifacts generated and the actors involved in these processes.
For authors, AI accelerates literature synthesis and manuscript drafting (actions), yields an exponential surge in submitted manuscripts (artifacts) and establishes itself as a collaborative partner to human researchers (co-actors). Conversely, for publishers, AI expedites peer review and editorial curation (actions), alters the nature of final publications (artifacts) and actively supports editors and referees (co-actors). This shifting actor landscape upends traditional conventions of authorship, responsibility and accountability. It calls for a deliberate redesign of the architecture of interactions among actors, actions and artifacts (the 3As) within the publishing ecosystem. To guide this strategic governance, the publishing community must resolve two critical inquiries.
First, shall we consider AI as a formal co-author? This question becomes unavoidable as AI achieves greater autonomy across the scientific research pipeline, from literature gap identification and hypothesis generation to autonomous experimentation and complete manuscript drafting (Gottweis, J., et al., 2026. Accelerating scientific discovery with co-scientist. Nature 655 (8122), 487–496.). The trajectory toward autonomous machine intelligence in science and discovery appears irreversible. While traditional frameworks treat AI strictly as an instrument, its evolution into an intelligent partner contributing synthetic reasoning and creativity profoundly challenges this paradigm. The boundary between tool and creator is blurring. Yet whether AI is ultimately recognized as an author, one principle is non-negotiable: the undisclosed use of generative AI in research production is fundamentally unethical. In adherence to this principle, I declare here that the viewpoints and initial draft of this editorial are entirely my own, whereas AI was utilized solely to proofread and refine the text.
Second, shall we acknowledge AI as a co-reviewer? As referees increasingly leverage AI to screen manuscripts, synthesize critiques and draft evaluation reports, the traditional peer review model faces unprecedented stress. In many instances, AI can generate more systematic, structured and rigorous paper assessments than overextended human experts or non-expert reviewers. This capability offers vital relief to editorial boards grappling with a widening gap between escalating submission volumes and reviewer availability. Nevertheless, this shift introduces complications regarding transparency, accountability and evaluative integrity. Institutional policies remain fragmented across publishers. Because peer review intrinsically hinges on confidentiality and trust, the undisclosed use of AI compromises the integrity of the evaluation process.
While definitive answers to these two questions remain elusive, they should compel publishers to intentionally redesign their 3A architectures: the author–drafting–manuscript subsystem and the reviewer–evaluating–report subsystem. Such redesign must balance innovation with responsible governance. As we transition toward human–AI hybrid publishing ecosystems, procedural ethics and integrity must serve as our uncompromised baseline. Ultimately, whether the operating actors are human, AI or hybrid, the ethical foundation of the publishing system must remain resolute.
Carolyn Seepersad
Providing guidance to authors on responsible and permissible use of AI
One of the most significant challenges facing journal editorial staff is providing guidance to authors on responsible and permissible use of AI and ML. Although it is tempting to forbid its use entirely – a path that some instructors take in the classroom – blanket prohibitions are not reasonable given the pervasiveness of AI in professional practice. Many engineering researchers use AI to provide basic editorial checks (e.g., spelling, grammar, usage), to enhance search engine responses to queries, to identify gaps in literature reviews, to accelerate scientific programming and as an important tool in modeling, simulation and AI-assisted design. Given this widespread use in a standard research setting, the question becomes how to establish guidelines that are effective and meaningful for authors.
One example of AI guidelines for authors is the position statement issued by ASME (https://www.asme.org/getmedia/22767fd2–0158-4feb-96ff-cbeb990b5b84/position-statement-use-of-ai-asme-external-content.pdf), which governs publications in ASME journals, including the Journal of Mechanical Design. The statement encourages authors to focus on authorship, transparency and reproducibility. In terms of transparency, any use of AI, such as editing or polishing text or providing feedback on research design, must be acknowledged explicitly in the manuscript. When a manuscript includes AI or ML algorithm development, transparency and reproducibility require the author to document the algorithm itself (version, source), how it was used, the data sets used to train it, the prompts used to generate results, the parameters used to tune the model and the procedure for evaluating its performance.
Aside from the permissible use of AI or ML as part of algorithm development for engineering applications, authors are not permitted to use AI or ML to generate substantive portions of their manuscripts. This guidance is tied to the issue of authorship, and is perhaps the most difficult to interpret. As the use of AI grows, new and seasoned authors have begun to raise questions about what counts as substantive portions of a manuscript.
For example, what if an author uses AI to help generate equations for analyzing results? Or to double-check calculations? Or to design an experiment or the steps of a research study? These questions echo long-term questions about human authorship, and one way for editors to respond is to invite authors to substitute a human colleague for the AI tool and ask whether they would include a human colleague as coauthor on the paper if that colleague performed the same tasks. If the answer is yes, then it is clearly a substantive contribution to the paper and is disallowed by most journals. If they would elect to include the colleague in the acknowledgments, then it may be a permissible use of AI that can be mentioned in the acknowledgments section or the text itself. However, as AI permeates more of our digital tools, it will become increasingly difficult to distinguish clearly between the role of the human and that of the AI assistant. Accordingly, it will become more and more difficult for publication guidance to maintain its relevance and currency.
Helping authors reason through the permissible use of AI, rather than providing highly specific and immutable guidelines, may be a better long-term approach in this dynamic research environment.
Concluding remarks
There was no effort to synthesize or draw conclusions from the panelists’ remarks, and readers can draw their own conclusions. Two points, however, are worth noting. First, publishing in the age of AI presents major challenges, and how research will be defined, conducted and communicated in the future is an open question; notably, all design journals embrace the same philosophy where the human and the artifact are inextricably connected – at least for now (the irresistible emdash included). Second, AI (re-) opens fundamental questions of what it means to be human and emphasizes the need for training in critical thinking and for design research that creates space for genuine human–human dialogue.
Open access
