The framework’s central insight rests on a distinction that cognitive load theory has refined over decades: not all mental effort is equal, and not all of it deserves to be automated. Intrinsic load refers to the inherent complexity of a task — the number of elements a writer must hold in mind simultaneously to make sense of an argument. Extraneous load, by contrast, is the friction that contributes nothing to understanding: hunting for a citation format, wrestling with a template, untangling a garbled bibliography. Germane processing is the deliberate allocation of working memory resources toward building schemas, the mental structures that make expertise possible. The paper’s argument is that AI tools are remarkably well suited to absorbing extraneous load and strikingly ill suited to absorbing germane processing, because germane processing is not a task that can be handed off at all. When a tool performs the work, the processing does not transfer to the writer; it simply does not occur.

This claim gains force from recent empirical findings on what happens when writers lean heavily on generative assistance. Studies documenting reduced neural engagement, weaker recall of one’s own writing, and a diminished sense of authorship among heavy AI users suggest that the costs of offloading are not merely pedagogical abstractions. They are measurable changes in how deeply a writer encodes their own ideas. Writing, as Janet Emig argued in 1977, is a distinct mode of learning rather than a record of thinking already completed. The act of struggling to render a messy idea in precise language is itself the mechanism by which the idea becomes clear to the writer. Remove the struggle, and the clarity never arrives.

The evidence on writing-to-learn adds an important nuance. A review of 48 school-based writing-to-learn programs by Bangert-Drowns and colleagues found a consistent positive effect on achievement, and a subsequent synthesis by Graham and colleagues reported comparable effects across 56 experiments in science, social studies, and mathematics. But what predicted a stronger effect is instructive for AI design: writing that included metacognitive prompts helped more, while assigning longer pieces helped less. The benefit of writing comes from the requirement that writers think about what they are doing, not from sheer output volume. This is precisely the condition that AI can silently remove. A student who prompts a chatbot to expand a paragraph has produced more words while doing less thinking — the exact inversion of what the writing-to-learn literature identifies as beneficial.

The nine-phase framework responds to this problem by refusing to treat research writing as a single undifferentiated activity. Instead, it maps the process from initial curiosity and inquiry through references and submission, and for each phase specifies three things: the dominant cognitive demand, the appropriate AI role ranging from minimal involvement to full collaboration, and design guidelines for how tools and interactions should behave. The logic is that the right AI role depends on the cognitive demand of the phase, not on the tool’s capabilities. In early phases, where the writer must generate questions and confront ambiguity, AI involvement should be minimal, because the struggle to formulate a researchable question is germane processing in its purest form. In later phases, where formatting conventions and citation management dominate, AI can operate at full collaboration, because those demands are largely extraneous and offloading them frees working memory for the thinking that matters.

This phase-aware approach is an application of responsive pedagogy to tool design. Responsive pedagogy holds that instruction should respond to the learner’s current understanding through recursive feedback dialog rather than proceeding on a fixed script. Transposed to AI-assisted writing, the principle becomes: the tool should respond to the cognitive demands of the task at hand, adapting its level of involvement as the writer moves through phases. A system that offers the same degree of assistance at every stage — drafting, organizing, synthesizing, polishing — is pedagogically blind, because it cannot distinguish between the friction that should be removed and the friction that is doing the teaching.

The framework’s scope is unusually broad. It addresses research writing from secondary school students in grades 9 through 12 developing their first researchable questions, through doctoral candidates building dissertations, to postdoctoral and professional researchers preparing manuscripts for peer review, across disciplines. This range matters because the temptation to offload struggle is not confined to novices. The paper opens with an anecdote about a high school student who submitted three AI-generated papers, frustrated by what the student perceived as a double standard — teachers use AI all the time, so why is it not acceptable for students? The anecdote illustrates how pervasive the temptation has become, and how the absence of clear norms leaves students improvising their own ethical boundaries. But the same dynamics apply to a seasoned researcher facing a deadline, for whom the pull toward delegation is arguably stronger because the procedural fluency to prompt effectively is already in place.

The paper’s own composition models its recommendations. Each section includes a human-AI collaboration disclaimer documenting exactly which cognitive work the researcher performed and which was delegated. The researcher conceived the paper’s focus, proposed the conceptual framework, and implemented it. Grammarly was used for editing, with sentence revision suggestions accepted or dismissed by the researcher. When Claude Pro was asked to review the initial single research question, it suggested splitting the question into two — one addressing which cognitive demands constitute extraneous load versus germane processing, the other addressing how human-in-the-loop collaboration should be designed across phases — and the researcher, after review, agreed. In building a definitions table, the researcher set the structure, identified terms, and selected sources; the AI drafted definitions grounded in those researcher-selected sources and flagged terms the researcher had overlooked. The researcher then verified each entry against the cited sources and reorganized the categories. This is the framework applied to itself: human curation and judgment retained, AI operating within boundaries the human establishes.

The transparency of these disclaimers points toward a broader implication for how AI-assisted scholarship should be evaluated. If the value of a paper lies partly in the germane processing its author performed, then disclosure of the division of cognitive labor becomes a form of methodological honesty, analogous to reporting data collection procedures. Readers can then judge for themselves whether the claims rest on understanding the author actually built or on synthesis the tool performed. The framework implies that assessment practices — in classrooms, in doctoral programs, in peer review — may need to shift from evaluating final products toward evaluating the process by which they were produced, since the product alone can no longer reveal whether the thinking occurred.

Equity considerations complicate the picture. The framework’s implications for responsive digital education include tool selection, assessment, and support for writers with cognitive differences. For some learners, certain demands that are extraneous for others may function differently, and rigid phase prescriptions could either burden or liberate them inappropriately. A student with dyslexia, for example, may experience transcription demands as genuinely extraneous friction that assistive tools should remove, while still needing to own the conceptual struggle of argument development. The framework’s emphasis on human judgment as the operational mechanism — rather than fixed rules about which tools may be used when — is what allows this flexibility. The line between offloadable and non-offloadable work is drawn by a person who understands the learner and the task, not by a policy document or a software default.

What emerges is a design philosophy rather than a prohibition. The paper does not argue that AI should be kept out of research writing; it argues that AI’s role should be calibrated to preserve the cognitive work that makes research meaningful. The productive struggle literature, drawing on work by Hiebert and Grouws and by Warshauer, positions struggle as a means of engaging in and developing critical thinking — not as an obstacle to learning but as its engine. Anne Lamott’s familiar advice to her overwhelmed brother facing an overdue report on birds — take it bird by bird — captures the framework’s practical spirit: large cognitive tasks become manageable when broken into steps, but the steps themselves must be climbed by the writer. AI can hold the ladder, hand over the field guide, and check the citations. What it cannot do is climb, because the climbing is the learning. The framework’s contribution is to specify, phase by phase, where the ladder belongs and where the writer’s feet must stay on the ground.

Subject of Research: A nine-phase human-in-the-loop framework for AI-assisted research writing based on cognitive load theory

Article Title: Designing a human-in-the-loop AI iterative process for research writing: a framework for preserving productive struggle

Article References: Chauncey, S. A. (2026). Designing a human-in-the-loop AI iterative process for research writing: a framework for preserving productive struggle. Frontiers in Computer Science, 8, Article 1947835. https://doi.org/10.3389/fcomp.2026.1947835

Image Credits: AI Generated

DOI: 10.3389/fcomp.2026.1947835

Keywords: artificial intelligence, research writing, cognitive load theory, productive struggle, human-in-the-loop, generative AI, education, academic writing, cognitive offloading, responsive pedagogy, NotebookLM, authorship

The editorial draws a careful distinction between two fundamentally different categories of AI use. The first involves non-generative tasks, such as data classification, pattern sorting, and copyediting. In these applications, AI functions much like a calculator or a statistical software package: it processes information according to defined procedures, and a human remains responsible for interpreting the results. The authors judge this kind of use to be broadly acceptable. It accelerates work that would otherwise consume countless hours, and it does so without displacing the judgment that makes research meaningful. A tool that cleans up prose or organizes a dataset does not decide what a study means or whether its conclusions are trustworthy.

The second category is far more troubling in the authors’ view. When researchers turn to generative AI to produce scientific ideas, seed research questions, or draft the interpretive substance of a manuscript, the machine stops being an instrument and starts being a substitute. The editorial argues that this kind of use deserves substantially more scrutiny than it currently receives, and that it should be disclosed in scientific manuscripts so that readers, reviewers, and the wider community can evaluate how much of the work originated in a human mind. Transparency, in this framework, is not a bureaucratic burden but a precondition for trust.

The authors go further than mere disclosure. They propose that certain uses of generative AI should be disallowed outright. Specifically, they identify the core intellectual, creative, ethical, interpretive, and accountability-bearing work of research as territory where AI must not replace human interpretation and judgment. A scientific paper is not merely a collection of sentences; it is a chain of responsibility that runs from the initial question through the gathering of evidence to the final claim about what that evidence shows. If any link in that chain is forged by a system that cannot be held accountable, the entire structure is weakened. Responsibility, the editorial insists, cannot be delegated to software.

Among the specific prohibitions the authors propose are some that may surprise researchers who have already begun experimenting with AI brainstorming tools. They argue that scientists should not use AI to generate their own initial research ideas or research questions. The origin of a study, they contend, should be a human act, rooted in curiosity, observation, and lived scientific experience. Likewise, they argue that ethical determinations, risk assessments, and policy recommendations should never be outsourced to generative systems. These are precisely the tasks where errors carry the greatest consequences for people, communities, and the environment, and where the absence of human accountability would be most damaging.

One of the most striking arguments in the editorial concerns serendipity, the happy accidents that have repeatedly driven scientific progress. The authors warn that if every researcher relies on the same AI tools with similar inputs, the diversity of thought that produces unexpected discoveries will shrink. Generative models are trained to produce probable outputs based on known patterns in existing data. They are, by design, engines of the average and the expected. Science, by contrast, has often advanced by departing from prior assumptions and breaking free of typical patterns. Penicillin, cosmic microwave background radiation, and countless other breakthroughs emerged not from the most probable answer but from an anomaly that a human noticed and refused to ignore. A research ecosystem optimized for probable outputs is an ecosystem primed to miss exactly those anomalies.

This concern leads the authors to a broader warning about optimization. Science, they argue, should not be unthinkingly optimized as if it were a manufacturing workflow producing standardized outputs. A factory benefits from consistency, repeatability, and the elimination of variation. Science requires the opposite: a wide diversity of inputs, approaches, and ideas. When the same models, trained on the same corpora, are applied across millions of research projects, the risk is a homogenization of the scientific imagination itself. Questions that fall outside the statistical grain of the training data become less likely to be asked at all, not because they are unimportant but because no human curiosity set them in motion.

The editorial does not call for a retreat from technology. Its authors are careful to preserve a role for AI that is genuinely augmentative. Non-generative applications, from classifying large datasets to polishing language, can free researchers to spend more time on the work that only they can do: conceiving questions, designing studies, weighing evidence, and taking responsibility for conclusions. The problem arises not from computation but from substitution, not from assistance but from abdication. In this sense, the proposed norms are less a restriction on AI than a defense of the human role that gives AI its value in the first place.

The authors distill their argument into a clear standard for the scientific enterprise. According to the editorial, the originating research question and the final evidentiary judgment in every scientific publication should be produced by human minds. Everything between those two poles may, with appropriate care and disclosure, involve machine assistance. But the beginning and the end of the scientific process, the spark of inquiry and the verdict on the evidence, must remain human. This framing offers journals, institutions, and funding agencies a practical benchmark against which policies on AI use can be measured, and it gives individual researchers a simple test to apply to their own workflows.

As generative AI becomes more capable and more deeply embedded in the research infrastructure, debates of this kind are likely to intensify. The PNAS Nexus editorial contributes a principled starting point: AI should amplify human science, not quietly absorb it. By distinguishing acceptable non-generative uses from the generative applications that demand scrutiny and disclosure, and by drawing firm lines around idea generation, ethical judgment, and accountability, Branas and Levine offer the scientific community a framework for preserving what makes research trustworthy. The message to researchers is unambiguous. Use the tools to work faster and see further, but never hand over the questions, the judgment, or the responsibility that define science itself.

Subject of Research: Norms for the appropriate use of artificial intelligence in scientific research

Article Title: Augmentation not abdication, when using AI in research

Article References: Augmentation not abdication, when using AI in research. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, generative AI, research ethics, scientific publishing, PNAS Nexus, authorship, serendipity, research norms, AI disclosure, human-led science, accountability, scientific method

The research, led by Md. Saiful Alam of the World University of Bangladesh together with colleagues from institutions in Bangladesh and Malaysia, frames the relationship between non-native English-speaking researchers and ChatGPT as an epistemic collaboration with a strict division of labor. The authors describe ChatGPT’s role as peripheral and non-authorial: the tool edits language, corrects grammar and punctuation, and suggests rewording, but it never generates the science or claims credit for the work. The human researcher, meanwhile, retains what the team calls optimal authorial anthropocentricity, meaning full ownership of the ideas, the content, and the final text.

The stakes are high. Non-native English speakers make up roughly 95 percent of the global population, and previous studies cited in the paper show they face delays, rejections, and significant financial costs when publishing in English-only journals. Journals often recommend paid language editing by native English speakers, a service that can burden researchers in the global South who frequently publish on their own budgets without institutional funding. The lead author recounts one such recommendation from a journal, which told him his article needed proofreading by a native English speaker to correct grammatical and syntactical issues reducing fluency.

To explore whether ChatGPT could offer an alternative, the team conducted a qualitative study grounded in a constructivist paradigm, treating the authors’ own lived experiences as data. They composed short research paper excerpts containing deliberate language issues, ran them through QuillBot’s free grammar checker to identify errors, and then asked the free version of ChatGPT 3.5 to rewrite the texts in correct English. The conversations were captured as images using Snip & Sketch and analyzed with adapted techniques from digital data mining and educational data mining, including text categorization, segmentation, and topic mapping.

The results were striking in their specificity. In one demonstration, a 31-word paragraph written by a non-native researcher contained five errors flagged by the grammar checker, involving word choice, prepositions, and punctuation. When the same text was given to ChatGPT with a simple request to rewrite it in correct English, the tool corrected every flagged error without adding or subtracting any content. Crucially, the authors note, ChatGPT did not create a new text; it revised the existing one, which means composition credit, and therefore authorship, remains with the human writer.

The study also documents where the tool falls short, and these limitations define the researcher’s responsibilities. ChatGPT does not reliably edit in-text citations to match APA conventions, sometimes expands abbreviations incorrectly, and can apply unrequested styling such as double quotation marks around well-known terms, which conflicts with APA norms. In one example, the tool misinterpreted the abbreviation M-SGTs as Multilingual Student-Generated Teachers when the authors had intended Multi-Subject Generalist Teachers. Because the model cannot see the entire paper or grasp its contextual nuances, the researchers argue, a careful human review of every AI-edited passage is essential before anything is incorporated into a manuscript.

Beyond mechanical fixes, the findings suggest ChatGPT can help researchers internalize the conventions of scientific writing itself. By comparing their original drafts with the edited versions, authors can notice patterns, such as the correct punctuation after the word including, and gradually master the tone, voice, and discursive style expected by high-profile journals. The team describes this as noticing-based grammar learning, a process that may reduce reliance on the tool over time and strengthen the researcher’s independent command of academic English.

The practical outcomes reported are notable. According to the paper, fifteen non-native English-speaking researchers applied the proposed workflow and subsequently published fifteen papers in English-only journals indexed in Scopus and Web of Science, without receiving editorial requests about language quality or being compelled to purchase native-speaker editing services. The authors report that the tool’s performance in correcting language errors reached a 99 percent accuracy rate in their demonstrations, and that the approach aligns with publisher policies, including statements from major platforms that generative AI is not an author and should only be used to improve language and readability with caution.

The team is careful to position the framework within ethical boundaries. Because ChatGPT’s participation is limited to language editing, the collaboration falls within what most scholars and journals consider ethical AI use, though the authors recommend disclosure that the tool provided digital labor for language purposes. The three principles underpinning their model are the pragmatization of the language problem, the deployment of ChatGPT’s digital instrumentality, and the retention of the researcher’s optimal authorial anthropocentrism, ensuring that creation, not mere generation, remains the defining feature of scientific knowledge production.

The authors acknowledge the study’s limitations: it rests on a small sample of personal, first-hand experiences and simulated conversation data rather than large-scale empirical evidence. They call for longitudinal, positivist studies to establish generalizability. Still, the paper offers a provocative vision for the future of scientific publishing, one in which a free AI tool serves as a peripheral editing partner, researchers from the global South gain equitable access to English-language journals, and the chronic linguistic gatekeeping of academic publishing begins to lose its grip.

Subject of Research: Using ChatGPT as a peripheral language-editing tool to support non-native English-speaking researchers' scientific writing

Article Title: Framing an epistemic collaboration between chat GPT’s peripheral instrumentality and the author’s optimal anthropocentrism for enhancing non-native-English-speaking researchers’ scientific writing

Article References: Alam, M. S., Asmawi, A., Haque, M. H., Fatema, S., Ullah, M. M., Patwary, M. N., Fatema, T., & Akhter, H. (2026). Framing an epistemic collaboration between chat GPT’s peripheral instrumentality and the author’s optimal anthropocentrism for enhancing non-native-English-speaking researchers’ scientific writing. Discover Education, 5(1), Article 1148. https://doi.org/10.1007/s44217-026-02263-6

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02263-6

Keywords: ChatGPT, generative AI, scientific writing, non-native English speakers, academic publishing, language editing, authorship, linguistic justice, Discover Education, epistemic collaboration, global South, research ethics

The concern driving calls for radical transparency is real. In a traditional workflow, named authors are presumed to have made a substantial intellectual contribution and to be able to answer for the integrity of their work. Large language models complicate that presumption because they can now help with brainstorming, structuring, rewriting, and even drafting entire manuscripts from a handful of instructions. Yet under prevailing guidelines, models cannot be listed as authors, since authorship is thought to require moral accountability that machines lack. The result is a gap: a researcher could, in principle, coax a minimally publishable paper out of an AI system, attach their name, and submit it without being detected, because the model will never object to being passed over.

The authors acknowledge that the problem may be worse than before the AI era. Prior to roughly 2023, anyone trying to claim credit for a paper they did not write had to contend with at least one other human collaborator, creating a natural disincentive. They cite a documented case, drawn from the secondary literature, of an academic producing articles on seemingly unrelated topics that may be almost entirely AI-generated. That author openly declares AI use and claims to have performed the editing, which technically satisfies current transparency standards, yet reviewers have no way of knowing whether one percent or one hundred percent of the text was machine-generated. The advocates of Mandatory Full Inclusion argue that requiring the raw receipts would let editors verify authorship claims directly, shifting the culture from blind trust in attestation statements toward verifiable trust based on evidence.

The new analysis does not dispute the motivation, but it dismantles the mechanism. The first and most obvious problem is unenforceability. A person unscrupulous enough to claim authorship on a largely AI-generated paper is, by hypothesis, someone willing to misrepresent their contribution, and such a person facing a choice between uploading materials that would immediately falsify their claim or simply not declaring AI use at all has strong incentives to choose the latter. AI detection software remains unreliable, and models can be prompted to scrub their own output of machine-like signatures. The likely outcome, the authors argue, is that unwarranted authorship claims would continue apace while honest declarations of AI use would decline, a result directly opposite to the policy’s intent.

Even for honest researchers, compliance may be technically infeasible. Serious scholarly work often involves consulting multiple models across many sessions, sometimes having the models critique each other’s responses to reduce the biases of any single training distribution. Built-in limits on context length, and the tendency of language models to underuse information in the middle of long interactions, give users further reasons to break work into separate exchanges. In collaborative projects where several co-authors each use one or more models over months, the resulting transcripts could run to hundreds or thousands of pages. Locating, exporting, redacting sensitive material, anonymizing for peer review, and assembling such a file could consume many hours of administrative labor that contributes nothing to the substance of the paper.

That burden creates what the authors call a transparency paradox, supported by recent empirical work. In a survey of twenty bioethics journals, only twelve papers out of more than one thousand published in 2024 referred to AI use or non-use, even though independent evidence suggests actual usage is many times higher. Experimental studies add that disclosed AI assistance can reduce perceived writing quality and trust relative to non-disclosure. A policy that attaches heavy documentation costs and potential stigma to every declaration would, the authors predict, push otherwise honest researchers toward omitting disclosure entirely, worsening the very under-reporting problem it aims to solve.

Perhaps the most novel objection concerns the Hawthorne effect, the well-documented tendency of people to change their behavior when they know they are being observed. If researchers know that every interaction with a model may be read by editors, reviewers, and readers, they may stop asking clarifying questions for fear of appearing under-informed, abandon creative back-and-forth for fear of seeming unserious, and avoid unconventional angles for fear of entertaining unpopular viewpoints. Evidence for this risk comes from a study of shared language model accounts, in which users reported that because conversational histories might be seen by others, they avoided personal or speculative topics, adopted alternative phrasings, deleted histories, or created separate accounts for certain queries. The authors argue that generative AI can serve as a private interlocutor during inquiry, a space for testing formulations, rehearsing objections, and exploring lines of thought that may later be abandoned. Mandatory surveillance of that space could shrink the exploratory thinking on which the best creative scholarship depends, and could produce a misleading kind of transparency: an artificially curated record shaped by the disclosure requirement itself rather than an accurate account of how ideas actually developed.

The equity costs fall unevenly. The authors identify four groups for whom language models can partly compensate for disadvantages that more privileged colleagues do not face: researchers whose first language is not English, who can use the tools to approach the fluency reviewers expect; researchers with conditions affecting writing, such as dyslexia, ADHD, chronic fatigue, repetitive strain injury, multiple sclerosis, or motor impairments; researchers with heavy caregiving responsibilities, who are disproportionately women and who may benefit from drafting in short bursts; and scholars at under-resourced institutions, particularly in the Global South, who often lack the postdocs, editorial assistants, and writing support available at well-funded universities. Suppressing AI use through surveillance-heavy documentation requirements could therefore widen existing disparities in who gets published, even if institutions must simultaneously pursue the structural reforms that address the root causes of those disparities.

The policy also asks the impossible of the people who would have to do the checking. Peer reviewers are selected for expertise in a manuscript’s subject matter, not for the ability to judge from a prompt transcript how much of a text reflects substantive human contribution versus generic machine generation, a skill no current training program cultivates. Reviewer recruitment is already a serious bottleneck, especially in the humanities and qualitative social sciences, where editors may struggle to find even one suitable expert per submission. The predictable result is that supplemental files would go unread or receive only superficial assessment, defeating the policy’s purpose. Moreover, the authors note, academic publishing has never demanded this level of documentation of anyone: researchers are not required to upload minutes of every meeting, every tracked-changes draft, or records of which paragraph belongs to which co-author, because the costs and chilling effects are judged to outweigh the benefits. Singling out AI use for demands far exceeding those imposed on human collaborators is inconsistent with the field’s established, and well-justified, balance between transparency and other goods.

In place of Mandatory Full Inclusion, the authors propose a proportionate framework. First, routine declaration of AI use in all manuscripts, whether or not AI was used, to normalize transparency; one proposed scheme offers three simple options: no generative AI use, AI limited to copyediting human-drafted text, or substantive AI use described narratively. Second, targeted methods-level disclosure, including prompt logs where necessary, when AI bears on scientific validity or reproducibility, such as in data analysis, code generation, or structured literature synthesis, consistent with emerging reporting checklists. Third, voluntary submission of prompts and outputs for authors who wish to document their own contribution, potentially through summary-level methods such as the SLICE approach, in which independent assessors estimate the percentage of human and AI input at each stage of writing without exposing raw transcripts. Finally, journals should retain and enforce the requirement that anyone claiming authorship explain how they meet the relevant criteria, through ICMJE-style attestations or CRediT statements, and treat false attestations as misconduct. The authors concede that a trust-plus-accountability system is imperfect and that some will exploit it, but they conclude that mandatory full prompt disclosure would not fix that problem and would instead introduce surveillance harms, discourage beneficial uses of AI, and disadvantage the very scholars who need these tools most, ultimately doing more harm than good.

Subject of Research: Ethics of mandatory disclosure of AI prompts in scholarly publishing

Article Title: Against mandatory prompt disclosure in AI-assisted scholarship

Article References: Earp, B. D., Schüklenk, U., Savulescu, J., & Porsdam Mann, S. (2026). Against mandatory prompt disclosure in AI-assisted scholarship. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03326-w

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03326-w

Keywords: large language models, academic publishing, research integrity, authorship, prompt disclosure, transparency, Hawthorne effect, peer review, AI ethics, scholarly writing, equity, research policy

The authors structure their argument around three classical ethical traditions: consequentialism, which judges actions by their outcomes; deontology, which judges them by duties and rules; and virtue ethics, which judges them by the character of the person performing them. Each lens, they contend, reveals a different facet of the same underlying problem. A consequentialist might ask whether prompt-generated text improves or degrades the collective quality of scientific knowledge. A deontologist might ask whether researchers have a duty to be honest about how their manuscripts came into being. A virtue ethicist might ask what kind of scholar one becomes when the intellectual labor of drafting is handed to a machine. The fact that all three traditions converge on skepticism, the authors suggest, is itself a signal worth taking seriously.

The consequentialist case is complicated by genuinely mixed evidence. Experimental work published in Science in 2023 by S. Noy and W. Zhang found that generative artificial intelligence tools can substantially raise productivity in writing tasks, and a large-scale survey of German researchers reported in Research Policy in 2026 documented how widely these tools have already penetrated academic workflows. Yet a 2026 study in Nature by Q. Hao and colleagues found a troubling trade-off: artificial intelligence tools may expand the individual impact of scientists while contracting the overall focus of science, homogenizing research agendas as everyone draws on the same models. Daly and Teixeira da Silva add their own contribution to this literature with a 2026 paper defining what they call academic slop, the flood of low-quality, machine-flavored text now circulating in scholarly venues, and they point to a Lancet audit of 2.5 million biomedical papers that documented the scale of fabricated citations as evidence that the downstream harms are not hypothetical.

The deontological perspective sharpens the issue of honesty and disclosure. If a researcher presents prompt-generated prose as their own writing, they arguably violate a duty of transparency to readers, reviewers, and the scientific record. The authors note that this duty is contested terrain: some ethicists, including M. Hosseini and colleagues writing in Research Ethics in 2025, have argued that disclosure of generative AI writing assistance should be voluntary, while others, such as G. Kendall in the Journal of Academic Ethics, have proposed that authors should include both the prompts and the generated text as part of any submission. A 2025 commentary in Nature Machine Intelligence by B. D. Earp and colleagues framed the deeper problem as one of provenance: once text can be generated on demand, readers can no longer trace ideas back to the minds that supposedly produced them. Daly and Teixeira da Silva side with the camp that sees disclosure as insufficient on its own, because disclosure does not answer the prior question of whether the practice should occur at all.

Virtue ethics supplies the emotional and cultural core of their argument. On this view, the point of academic writing was never merely to produce text; it was to cultivate the intellectual virtues of patience, precision, and honest engagement with evidence. Writing a first draft is how a scientist discovers what they actually think, confronts the gaps in their reasoning, and earns the right to put their name on a manuscript. Offloading that task to a language model, the authors argue, is not a neutral efficiency gain but a loss of practice, in the same sense that a craft tradition is lost when artisans stop making things by hand. Daly has developed this theme elsewhere, calling in a 2026 paper for a kind of arts and crafts movement in the university, and in another article for a low-tech academic virtue ethics suited to the age of generative AI. The philosopher A. Ferdman has made a parallel argument in Philosophy and Technology, contending that some practices are worth preserving precisely because knowing when not to offload a task to technology is itself a skill.

Central to the letter is a pointed question: are prompters equivalent to authors, and should they be rewarded as such? The authors answer that prompting, however skillful, is not the same activity as drafting. A prompt is a request; a draft is a commitment. The person who writes a first draft takes responsibility for every claim, every citation, and every nuance of argument, while the person who prompts a model delegates that responsibility to a system that cannot bear it. This distinction matters because authorship in science is not merely a credit line; it is a guarantee that named individuals stand behind the work, can defend it, and will answer for its failures. A 2026 paper in Nature Reviews Bioengineering by Earp and colleagues compressed the point into a memorable phrase: thinking is not only writing, but writing is how scientists think.

The authors anticipate a powerful objection. Insisting on human drafting, critics say, would disadvantage researchers who do not speak English as a first language, since language models can flatten linguistic barriers that have long excluded talented scientists from Anglophone journals. Daly and Teixeira da Silva respond in a note to their letter that this concern has been overtaken by technology: drafting in one’s native language and then translating to English, whether through DeepL or a chatbot, is now an instantaneous, one-click process. The choice, they argue, is therefore not between human writing and global participation, but between writing one’s own draft in any language and outsourcing the intellectual work itself. Translation assistance and idea generation are, on their account, morally distinct from having a machine compose the substance of a manuscript.

The letter also engages with the practical messiness of enforcement. AI detection tools have proven unreliable, with a 2026 analysis by P. Tsigaris and Teixeira da Silva in Next Research arguing that most AI detector findings are false positives or false negatives, a conclusion with serious implications for researchers wrongly accused of misconduct. The authors themselves have previously argued in these pages that detection is the responsibility of editors and publishers, and their new letter implicitly shifts the burden away from detection and toward prevention: if the culture of academia reaffirms the value of the human draft, the arms race between detectors and generators becomes less central. They also cite the GAIDeT taxonomy, developed by Y. Suchikova, N. Tsybuliak, Teixeira da Silva, and S. Nazarovets, which offers researchers a structured way to decide which tasks may legitimately be delegated to generative AI and which must remain human.

What emerges from the letter is not a Luddite manifesto but a defense of a specific craft at a specific moment of vulnerability. The authors acknowledge that large language models can simplify knowledge creation and that human authors who use them are not villains; the question they press is whether the reward structures of academia, in which authorship confers credit, careers, and trust, should extend to those whose primary contribution is a well-crafted prompt. Their answer preserves a line that many journals and funding agencies are still struggling to draw. Notably, the letter itself carries a declaration that no AI was used by either author for conceptualization, drafting, editing, or validation, a small but pointed demonstration that the practice they defend remains entirely feasible. As generative models grow more capable and more embedded in every stage of research, the letter stands as a reminder that the deepest questions they raise are not technical at all, but ancient ones about outcomes, duties, and the kind of people scholars choose to be.

Subject of Research: The ethics of prompting large language models in academic publishing

Article Title: The Ethics of Prompting Large Language Models in Academic Publishing

Article References: Daly, T., & da Silva, J. A. T. (2026). The Ethics of Prompting Large Language Models in Academic Publishing. Annals of Biomedical Engineering. https://doi.org/10.1007/s10439-026-04356-7

Image Credits: AI Generated

DOI: 10.1007/s10439-026-04356-7

Keywords: large language models, academic publishing, prompt engineering, research ethics, consequentialism, deontology, virtue ethics, authorship, generative AI, academic integrity, scientific writing, AI disclosure

The historical core of the argument rests on a deceptively simple idea: that an artwork can be a set of instructions. Conceptual art, which crystallized as a movement in the 1960s, prioritized the idea behind a work over the finished object. Sol LeWitt famously declared that the concept is central while execution is merely a perfunctory affair, and that the idea becomes a machine that makes the art. Around the same time, artists associated with the Fluxus collective, including John Cage, George Brecht, La Monte Young and Yoko Ono, developed event scores: brief, colloquial instructional statements intended to be performed by anyone, anywhere. Because these text-based pieces could be executed by any participant, they opened the artwork to endless interpretation and variation, effectively distributing the act of creation beyond the named artist.

Xie and Wong show how this instructional, rule-driven logic prefigured generative and algorithmic art. Citing Philip Galanter’s definition of generative art, in which the artist cedes partial or full control to an external system such as an algorithm, the authors note that the legitimacy of such systems was grounded in foundations laid by conceptual art. Fluxus founder George Maciunas argued that the artist need only create a concept or method, after which the form can be created independently of him. In the 1960s, the Japanese avant-garde artist Yamaguchi even supplied manufacturing instructions to factories to produce his works. What has changed, the study contends, is the participatory capacity of the external system: where assistants and factories once carried out the concept, generative AI now does so at unprecedented scale and speed.

The technical trajectory the authors describe is recent and rapid. Text-guided digital image generation advanced significantly with the introduction of generative adversarial networks in 2014, followed by Google’s Deep Dream in 2015, which hinted at the possibility of rendering conceptual directives in visual form. A watershed came in January 2021 with OpenAI’s release of CLIP, which demonstrated an unprecedented capacity for associative retrieval between textual and pictorial modalities. These techniques were subsequently integrated into diffusion models optimized for synthesizing images from semantically anchored descriptions. The upshot is that users with little knowledge of deep learning pipelines can now outsource image-making entirely to AI assistants configured through minimal language, a capability that reshapes who, or what, can complete an artwork.

This is where the study raises its sharpest concern. Yoko Ono described her instruction paintings as unfinished works, recasting art from a finished object into an unfinished process to be completed by others, and she sought to erase the sacred distance between object and participant. Audience participation became a hallmark of contemporary art, theorized by Claire Bishop as a breakdown of the artist as autonomous producer and by Nicolas Bourriaud as relational aesthetics, in which art happens in the relations between elements, human and non-human alike. But when a generative model is trained on Ono’s instructions and produces visual outputs on demand, the human participant’s substantive engagement with the underlying concept can become virtually non-existent. The participant risks becoming a passive provider of impetus, an incidental vessel transmitting directives rather than an active determiner of conceptual outcomes, reducing participation to a hollow illusion of democratic inclusion.

An instructive counterexample comes from the German artist Mario Klingemann, whose work Appropriate Response trained a model on 60,000 internet quotes to generate aphorisms, delivered to kneeling audience members through a prayerful trigger device. Because people interpreted the generated sentences against their own experiences, the machine’s outputs functioned like fortune-telling or divination. The authors observe that the roles of human and machine in this feedback loop are reversed relative to Ono’s practice: the artist’s trained model now occupies the position Ono once held, issuing instructions to the audience. More troubling, they argue, the false contingency created by computation bestows a quasi-divine authority on the machine that no human artist ever commanded, since audiences and artists always knew they belonged to the same species.

To escape this impasse, the paper turns to posthumanist theory. Drawing on Stefan Herbrechter’s emphasis on the biological and microbiological entanglement with nonhumans, and on the insight from the Human Genome Project that humans, like all organisms, can be understood as encoded informational entities, the authors argue for positioning humans and machines as equal conversational partners and co-creators of meaning. In this reconfigured triangular framework of artist, audience and technological mediator, the AI system becomes a creative agency serving both artist and audience, performing tasks that were previously theirs alone. Slovenian philosopher Slavoj Žižek’s description of the uncanny experience of the human mind directly integrated into a machine underscores why new participatory frameworks must abandon the illusion of the autonomy of personhood.

The study also distinguishes carefully between two divergent conceptual legacies that carry different implications for AI art. LeWitt’s wall paintings required assistants to physically execute his written instructions at specific times and places, with results varying according to who carried them out; his linear trajectory from conception to completion suggests that AI image generation legitimately realizes intentionally seeded ideas. Ono, by contrast, held that mental realization alone constitutes art, challenging any strict correlation between computational output and original conceptual intention; from her perspective, AI finalization of imagined works is a redundant, even counterintuitive step. The authors point to Casey Reas’s Software Structures from 2004, in which programmers translated LeWitt’s instructions into code that generated wall drawings, as an early demonstration that machine systems could substitute for human draftsmen while contributing their own contingency factors to the outcome.

Where, then, does human creativity reside? Using Nelson Goodman’s distinction between autographic works, which are singular and authenticity-based like paintings, and allographic works, which are repeatable and concept-based like scores, and Kwastek’s analysis of interactive art as an autographic technical system built on constitutive rules, the authors argue that in text-to-image generation human ingenuity lies in the conceptual components embodied in language, not in the final digital artifact. The human mission is no longer to output a finite art object but to complete a framework of constitutive rules. Prompt engineering becomes the crucial site of this creativity: humans adaptively craft prompts that leverage neural network training and syntax modulation, and researcher Oppenlaender characterizes the process as iterative refinement, haltable by the human collaborator at any stage. Because current text-to-image outputs remain imperfect and often require post-production, final outcomes still depend on supplemental human judgment rather than full computational autonomy.

The paper closes by reconciling these tensions. Rather than negating conceptual art’s dematerialized impulses, generative AI signals their expanded reformulation within a technoculture of co-authorship between human and computational domains, echoing Jean Tinguely’s 1950s Méta-matic drawing machines, which satirized human veneration of machines even as they prefigured automated creation. By revisiting the conceptual art legacy and its insistence on ideas over objects, Xie and Wong contend, artists and theorists can build frameworks that support emergent AI practices and drive a continued rethinking of posthuman creative partnerships, ensuring that the delegation of agency to machines enriches rather than hollows out the human act of making meaning.

Subject of Research: The relationship between 1960s conceptual art and generative AI art practices, focusing on authorship, participation and posthuman creativity

Article Title: From instructions to intelligent synthesis: conceptual art, distributed creativity, and the posthuman turn in AI-powered artistic practices

Article References: Xie, Y., & Wong, A. L. (2026). From instructions to intelligent synthesis: conceptual art, distributed creativity, and the posthuman turn in AI-powered artistic practices. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03316-y

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03316-y

Keywords: conceptual art, generative AI, Yoko Ono, Sol LeWitt, text-to-image generation, prompt engineering, posthumanism, participatory art, authorship, distributed creativity, digital art, AI & Society

Lam’s argument responds to a perspective by Dr. Andrew Lee, who observed that despite documentation being fundamental to psychiatric practice, formal instruction in how to do it remains surprisingly limited. Lam proposes that this gap reflects a deeper oversight in psychiatric education. If writing the note is the specialty’s defining procedure, then the neglect of explicit teaching around it amounts to overlooking one of psychiatry’s own central skills, the very act through which psychiatric reasoning becomes concrete, transferable, and accountable.

The technical core of the argument is a distinction between transcription and construction. Psychiatric assessment, Lam writes, drawing on enactive accounts of social cognition from researchers such as Hanne De Jaegher, Ezequiel Di Paolo, and Sanneke de Haan, is not simply the collection of subjective and objective data. It is an interpersonal process through which clinician and patient progressively make sense of the patient’s experiences together. Through dialogue, observation, collateral information, and reflection, disparate elements gradually coalesce into a coherent understanding that neither participant possessed at the start of the encounter. Documentation, then, is not the mechanical recording of what happened. It is the disciplined act of rendering that shared understanding into language.

This distinction matters because documentation is not merely the product of psychiatric reasoning. It is one of the principal ways that reasoning is performed. Every sentence in a note requires judgments about salience, context, interpretation, and meaning. The psychiatrist decides what belongs in the foreground, what remains in the background, how apparently unrelated observations fit together, and why one formulation better explains the patient’s presentation than another. Writing transforms tacit understanding into explicit clinical judgment. Thinking shapes writing, but writing, in turn, disciplines thinking, in a feedback loop that makes the note an instrument of cognition rather than a passive record of it.

From this perspective, the quality of psychiatric documentation cannot be judged by length. A well-written mental status examination, Lam argues, should allow another psychiatrist to almost reconstruct the encounter: to hear the cadence of the patient’s speech, follow the organization of thought, sense the emotional atmosphere, and appreciate the patient’s interpersonal style. The formulation should not merely catalogue diagnoses but reveal the reasoning that connects history, examination, differential diagnosis, and management. The goal is neither exhaustive description nor literary prose, but the faithful construction of a clinically meaningful narrative. Excellent documentation is distinguished not by verbosity but by intentionality, with every sentence deepening understanding.

The argument acquires particular urgency as artificial intelligence enters the clinic. Ambient scribes and generative AI tools are increasingly deployed to draft clinical notes automatically, promising to reduce documentation burden and improve efficiency. But Lam contends that these tools redefine what it means to author a psychiatric note. The central question is no longer who generated the words, but whether the psychiatrist intentionally endorses the clinical narrative those words express. If documentation is regarded merely as transcription, AI appears to offer a clean technological solution. If documentation is understood as the technical expression of clinical narrative construction, AI can assist only with the execution, never with the substance.

The reason is that the tasks AI cannot perform are not linguistic but clinical. An algorithm may summarize conversations, organize information, and render polished prose. It cannot determine which observations are clinically salient, distinguish signal from noise, weigh competing diagnostic formulations, appreciate the interpersonal dynamics through which understanding emerged, or assume responsibility for the interpretive judgments embedded in the final narrative. These capacities arise from the psychiatrist’s participation in the encounter and from the professional obligation to transform that encounter into coherent clinical meaning. No language model attends the appointment; therefore no language model can author its meaning.

On this account, the psychiatrist’s role evolves rather than diminishes. Expertise will increasingly be reflected less in the mechanics of writing than in the capacity to critically evaluate and refine AI-facilitated documentation, ensuring that it faithfully expresses the clinical narrative constructed during the encounter. The responsibility is not merely to verify factual accuracy or polish style, but to determine whether the note represents the understanding that actually emerged in the room. Authorship becomes an act of judgment rather than composition. The psychiatrist remains responsible not for editing a document, but for authoring its meaning.

The framework also carries concrete implications for medical education. If documentation is the technical execution of a defining procedure, Lam argues, it should be taught with the same intentionality and rigor afforded to other core psychiatric competencies. Rather than relying primarily on apprenticeship and informal observation, residency programs should provide explicit instruction in clinical narrative construction, deliberate opportunities for practice, structured feedback, and longitudinal assessment. A recent systematic review by Emekli and colleagues identified educational interventions, including workshops, guided writing, and individualized feedback, that improve documentation skills, while highlighting the need for more rigorous educational research in this area.

Residents, in other words, should learn not only how to document accurately, but how to determine what is clinically salient, synthesize complex and often conflicting information, justify diagnostic and therapeutic reasoning, and communicate that reasoning through a coherent narrative. These are procedural skills that demand deliberate practice and expert feedback rather than passive exposure. As the mechanics of writing become more automated, Lam concludes, training should place even greater emphasis on the distinctly human capacities that AI cannot replace: clinical reasoning, narrative synthesis, interpretive judgment, and the ability to construct meaningful understanding from the complexity of a psychiatric encounter. In an era increasingly shaped by artificial intelligence, the future of psychiatric documentation will depend less upon who writes the words than upon who authors the meaning.

Subject of Research: The role of clinical narrative construction in psychiatric documentation, education, and AI-assisted note writing

Article Title: Psychiatric Documentation as Clinical Narrative Construction

Article References: Lam, A. (2026). Psychiatric Documentation as Clinical Narrative Construction. Academic Psychiatry. https://doi.org/10.1007/s40596-026-02425-6

Image Credits: AI Generated

DOI: 10.1007/s40596-026-02425-6

Keywords: psychiatry, clinical documentation, medical education, artificial intelligence, clinical narrative, residency training, mental status examination, generative AI, clinical reasoning, participatory sense-making, Academic Psychiatry, authorship

The study, conducted by Iván Sánchez-López and Gemma San Cornelio of the Universitat Oberta de Catalunya in Barcelona together with Heleny Méndiz-Rojas of the Catholic University of the North in Chile, took a deliberately targeted approach. Rather than attempting an exhaustive census of the literature, the team mined Scopus and Web of Science for publications explicitly engaging both artificial intelligence and creativity, screened them through a PRISMA-informed procedure, de-duplicated the results down to 71 unique records, and then ranked them by citation prominence. The final corpus comprised the 40 most citation-prominent studies, spanning journal articles and full conference papers, all published between 2020 and 5 February 2025.

Once the corpus was assembled, the researchers analyzed it in ATLAS.ti using Grounded Theory and the Constant Comparative Method. Initial coding produced 227 descriptive code entries, refined into 203 first-order codes, which axial coding then organized into nine categories through 230 code-category assignments. A final narrative synthesis distilled eight cross-cutting transformation features. The team used the software’s AI-assisted coding only as a supplementary first pass; every machine suggestion was manually checked, and the final codes, categories, and interpretations were settled by the researchers themselves.

The central finding is that creativity, as portrayed across this influential literature, has become a socio-technical assemblage rather than an individual capacity. Creative outcomes now emerge from the interplay of human-model interaction, institutional gatekeeping, infrastructural asymmetries, and normative conflict. Generative AI is consistently associated with expanded productivity: faster ideation, rapid prototyping, broader exploration of possibility spaces, and new formats of human-AI collaboration. But the same studies repeatedly document a parallel intensification of trouble around authorship, originality, recognition, labour, bias, access, legitimacy, and sustainability.

The reorganization of creative work follows a recognizable pattern. As the cost of producing drafts and variants collapses, the bottleneck migrates from generation to framing, prompting, selecting, editing, and justifying outputs. In human-computer interaction and design research, this appears as iterative prompt craft, co-creative interfaces, and curation as the central creative acts. In education, generative tools are framed simultaneously as scaffolds for writing, feedback, and idea generation, and as threats to assessment, academic integrity, and critical thinking. In management and innovation studies, the emphasis falls on productivity and capability building, with experimental and empirical work showing that augmentation outcomes depend heavily on task conditions, organizational resources, and workflow integration rather than unfolding as a uniform boost to creativity.

The review argues that these expansions do not distribute their benefits evenly. Generative tools may lower the barriers to producing plausible artifacts, yet sector reviews suggest they can concentrate advantage around computing infrastructure, technical expertise, and validation authority. Barriers shift from production toward evaluation, credibility, and institutional endorsement, and access to tools does not automatically translate into recognition or effective use. The review also identifies a mismatch between productive acceleration and evaluative capacity: when systems flood the pipeline with candidates, criteria for quality, originality, and appropriateness do not scale automatically, and selection can drift toward popularity defaults, raising the risk of homogenization even as output volume rises.

Questions of evaluation are further complicated by evidence of human bias against machine-made work. Studies in the corpus position evaluators as gatekeepers whose identity-based perceptions of whether an artifact is human- or AI-produced shape recognition and uptake. In Csikszentmihalyi’s systems view of creativity, creative status is conferred by a field of evaluators, and generative AI disrupts that field function by multiplying and obscuring the sites of evaluation: merit, originality, and responsibility are now negotiated across platforms, curators, clients, reviewers, and model providers, while institutions cling to accountability logics designed for individualized authorship.

Reading the corpus against contemporary scholarship exposes two significant gaps. First, sustainability receives comparatively thin treatment. While the reviewed studies address it mainly through urban and infrastructural perspectives, recent research has developed systematic accounts of the material externalities of large-scale generative systems, including electricity demand, carbon and water footprints, data-centre infrastructure, and disclosure obligations. Second, Global South and minoritized perspectives remain weakly integrated into the dominant corpus, even though training data distributions, dominant-language resources, and uneven access to computation may act as technical and cultural priors that determine which outputs and epistemic positions become visible in the first place.

The authors are candid about the limits of their design. Focusing on the most-cited publications maps dominant framings but may reproduce existing epistemic hierarchies, since citational prominence is treated as an indicator of visibility and influence rather than quality or truth. The corpus also reflects indexed, anglophone, predominantly Global North academic infrastructures, and the 2025 slice of coverage is partial by construction. The researchers call for future work using ethnographies of everyday use, comparative studies across languages and cultures, and mixed methods connecting computational traceability with critical analysis of labour and environmental costs.

The study’s principal interpretive conclusion is that expanded productivity and contested authority are not separate outcomes but interrelated dimensions of AI-mediated creativity. Artificial intelligence does not simply automate creative tasks; it relocates where creative agency sits, changes how contribution is evaluated, and reshapes which actors and infrastructures participate in producing legitimacy. High-impact scholarship, the authors argue, does not merely describe this transformation. It also helps stabilize influential definitions of creative value and redistribute forms of recognition, making it all the more important to examine how authority, visibility, and material conditions determine which forms of AI-mediated creativity become visible and legitimate.

Subject of Research: A citation-informed review of how highly cited research frames the transformation of creativity under artificial intelligence

Article Title: Expanded productivity, contested authority: creativity and AI in high-impact research

Article References: Sánchez-López, I., Méndiz-Rojas, H., & San Cornelio, G. (2026). Expanded productivity, contested authority: creativity and AI in high-impact research. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03213-4

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03213-4

Keywords: artificial intelligence, creativity, generative AI, authorship, human-AI collaboration, creative industries, evaluation, legitimacy, bias, sustainability, research review, AI & Society

The concern emerges against a backdrop of rapid institutional response to generative AI. As large language models have become widely available, many journals, universities, and professional bodies have adopted policies that limit or restrict the use of AI in academic writing. Enforcement of these policies has leaned heavily on AI detection software, tools that attempt to distinguish machine-generated prose from human writing by analyzing stylistic patterns, vocabulary choices, and overall cadence. The problem, as Smith and Aggarwal emphasize, is that these tools are far from infallible. Citing recent research, they note that even automated detectors do not achieve perfect accuracy, and that human readers perform considerably worse. One study found that people cannot identify AI-generated writing any more reliably than random chance would allow.

Despite these limitations, anxiety about being accused of using AI is demonstrably shaping how authors write. The em dash has been singled out in public discourse, including commentary in major newspapers, as a marker of perceived overuse by AI tools. Yet the authors point out that peer-reviewed studies establishing a genuine association between the punctuation mark and machine generation are lacking. The association is largely a matter of perception, amplified by social media conversation and anecdote rather than systematic evidence. Still, perception can be powerful. Because the em dash may now carry a negative connotation, authors who wish to avoid evoking skepticism from reviewers and editors may simply stop using it.

The pressure is not hypothetical. In preparing a prior manuscript, one of the authors was advised by a mentor that the use of em dashes might be interpreted by reviewers as a sign of AI involvement. Both authors report hearing similar accounts from colleagues across academic medicine. These anecdotes suggest that a quiet recalibration of writing style is already underway, driven not by journal policy or formal guidance but by fear of suspicion. The authors argue that trainees and early-career academics, who are still developing their scholarly voices, may be particularly vulnerable to these pressures, internalizing the idea that certain stylistic choices are risky before they have had the chance to make those choices their own.

What is at stake, the correspondence contends, extends well beyond a preference for one punctuation mark over another. Writing, like other forms of art, is a medium through which individuals express personality, share perspective, and communicate thought. Stylistic signatures, including rhythm, cadence, and punctuation habits, are part of what makes an author’s voice recognizable. By unnecessarily restricting these elements, the academic community risks constricting what scholars are able to say and how they are able to connect with readers. The authors suggest this may be particularly damaging in fields such as psychiatry, where narrative skill and clarity of expression are not decorative extras but essential professional tools, central to both clinical communication and scholarship.

The technical reality of AI detection undercuts the logic of policing individual stylistic features. As Smith and Aggarwal observe, even automated detection software does not rely on any single tell in isolation. Instead, such systems perform a holistic analysis of a piece of writing, weighing overall style, paragraph length, grammatical choices, and vocabulary together before rendering a judgment. If the most sophisticated detectors require whole-text analysis to reach conclusions of imperfect reliability, then treating a lone punctuation mark as evidence of machine authorship is analytically indefensible. Overemphasis on a single element, such as punctuation considered in isolation, raises the very real potential for false accusations against human authors whose only offense is a fondness for the em dash.

The authors’ proposed alternative is straightforward: rather than asking authors to change how they write, the field should emphasize and demand transparency about AI use. This means clear authorship guidelines and explicit journal policies governing the use of generative AI, together with education for scholars on the importance of disclosure. Transparency, in this framing, addresses the actual problem, which is undisclosed machine assistance, rather than punishing stylistic expression that happens to overlap with the output of language models. The approach aligns with emerging guidance in medical education publishing, including recent international guidance on when and how to disclose AI use in academic publishing.

At the same time, the correspondence does not dismiss the need for safeguards. Smith and Aggarwal acknowledge that protections surrounding AI use are vitally necessary to prevent the field from becoming over-reliant on technology, with the risks to originality, accountability, and scientific integrity that such reliance would entail. Their argument is one of proportion: safeguards should target disclosure and integrity, not the erosion of individual voice. The distinction matters because style-based policing is both ineffective, given the weakness of the evidence linking any single feature to AI, and corrosive, given that it teaches a generation of writers to flatten their prose into something inoffensive and anonymous.

The em dash controversy is a small episode with outsized symbolic weight. It captures a moment in which institutions, reviewers, and writers themselves are groping toward norms for a technology that arrived faster than the rules governing it. The message from these psychiatry educators is a caution against premature judgment: with imperfect detection software, even less accurate human judgment, and only a weak and largely anecdotal association between a punctuation mark and machine text, the field should not allow a perceived association to change how people write. How we communicate, they argue, is too important to be dictated by suspicion of a dash.

Subject of Research: The impact of perceived AI-associated writing styles on academic authorship and publishing

Article Title: The Em Dash—Should We Allow AI to Change Our Writing Styles?

Article References: Smith, A. C., & Aggarwal, R. (2026). The Em Dash—Should We Allow AI to Change Our Writing Styles?. Academic Psychiatry. https://doi.org/10.1007/s40596-026-02435-4

Image Credits: AI Generated

DOI: 10.1007/s40596-026-02435-4

Keywords: em dash, artificial intelligence, academic writing, AI detection, writing style, academic publishing, authorship, transparency, psychiatry education, generative AI, scholarly voice, peer review