Scholarly Journals’ Editorial Policy Towards the Use of Artificial Intelligence in Social Sciences
Edited Transcript of Presentations
25 June 2026
Report compiled by Trishala Sancheti, Research Fellow, India Foundation
Table of Contents
1. Artificial Intelligence-Generated Text as a Challenge for a Scholarly Journal in the Field of Humanities……………………………………. …………………………………………………………………………….3
Prof. Lilia Kamolova
2. Transforming Governance Through Artificial Intelligence…………………………………………6
Wing Commander S. Sudhakaran (Retd)
3. Economic Cost Shifting of AI in Scholarly Publishing………………………………………………11
Prof. Alexey Kuznetsov (INION RAN)
4. AI’s Potential vs. Human Insight: Ethical Anxiety for Scholarship and Originality…..15
Prof. Sunaina Singh (India Foundation)
5. Stigmatised Artificial Intelligence Use: AI-Friendly Journal Policies vs. Quasi-Sensorial Editorial Practices………………………………………… ……………………………………………………………18
Sergey Gonov (INION RAN)
6. Formulating Editorial Standards and Addressing Structural Injustices in AI Governance23
Prof. Sushma Yadav
7. Reconceptualisation of Scientific Information and Book Culture in the Age of Artificial Intelligence………………………………………………… ………………………………………………………………27
Prof. Igor Buryanov (INION RAN)
8. Closing Remarks…………………………………………………………………………………………………….29
Maj. Gen. Dhruv C. Katoch and Prof. Alexey Kuznetso
Artificial Intelligence-Generated Text as a Challenge for a Scholarly Journal in the Field of Humanities
Prof. Lilia Kamolova
It is an honour to open the first session of this joint seminar on research integrity in AI generated academic papers and on the editorial policies of scholarly journals regarding the use of artificial intelligence. I hope our discussions today will help us address these complex issues, share experiences across academic systems, and develop practical mechanisms for handling manuscripts that incorporate AI-generated content. Drawing on my ten years’ experience as Editor-in-Chief of a scholarly journal, I would like to share the challenges academic publishers—particularly journals in the humanities—face today.
I will begin with a brief introduction to INION as an academic publisher. I will then explain what distinguishes a scholarly journal from a non-academic publication and highlight some of the unique characteristics of research publishing in the humanities. Finally, I will discuss the challenges posed by the rapid growth of AI-generated text and how these developments are affecting the publication of our own journals. INION conducts research across a broad spectrum of the social sciences and humanities, including political science, international relations, law, area studies, history, cultural studies, economics, sociology, religious studies, linguistics, literature, philosophy, library science, and related disciplines.
Alongside its research activities, INION is one of Russia’s leading publishers of scholarly literature in the social sciences and humanities. Each year, our publishing centre produces 24 academic journals and about 30 scholarly monographs and edited volumes. Four of our journals are indexed in the Web of Science and/or ERIH PLUS, sixteen are included in the Russian unified state list of scientific publications, and all are indexed in the Russian Science Citation Index.
Personally, I serve as Editor-in-Chief of a multidisciplinary journal and as Deputy Editor-in-Chief of the specialised journal Ethnopsycholinguistics. For our colleagues from India, I should add a brief note on the Russian publishing system. Most scholarly journals in the Russian Federation are published by government research institutes or universities. Publication in INION journals is free for authors, reviewers, and editorial board members. All publication costs are borne by INION, and every article is available on both the journal websites and the INION website, ensuring free access for readers.
Before discussing AI-generated manuscripts, it is important to clarify what we mean by a scholarly journal. The primary purpose of a scholarly journal is to advance knowledge through original research grounded in verified sources and conducted in accordance with recognised scientific methods. Scholarly articles present new findings, follow a structured academic format that enables readers to understand the research process from the formulation of the research question to the conclusions, and undergo rigorous peer review by experts in the relevant discipline.
Publishing in the humanities, however, differs in important ways from publishing in many other scientific fields. Research in the humanities relies predominantly on qualitative methods rather than quantitative analysis. Descriptive research is generally more common than experimental or modelling approaches. Consequently, the quality of the research depends heavily on the author’s expertise, analytical abilities, and deep subject knowledge. This context is particularly important when considering the impact of generative artificial intelligence. From a publisher’s perspective, the consequences can be summarised in several key observations.
First, AI has significantly lowered barriers for early-career researchers, including postgraduate students and recently appointed postdoctoral scholars. AI tools assist with literature searches, summarising sources, language editing, formatting, and bibliography preparation. As a result, many more submissions now pass the initial editorial screening because they are well written and technically compliant with journal requirements. While this improves accessibility for authors, it also creates additional work for publishers. Editorial offices receive a larger volume of manuscripts requiring expert assessment, increasing the burden on reviewers and often necessitating the expansion of reviewer pools. It may also require journals to introduce additional editorial screening stages before manuscripts proceed to peer review.
Second, the traditional concept of originality is increasingly difficult to apply. Generative AI challenges our understanding of what constitutes an author’s genuine intellectual contribution. A manuscript may contain entirely original wording yet contribute little or no genuinely new knowledge. Likewise, an author’s ability to craft sophisticated prompts does not necessarily reflect their research competence or analytical ability. Given the characteristics of humanities research, AI-generated content can appear in every section of a manuscript—from the title and abstract to the literature review, discussion, and even the references.
This raises an important question. Rather than simply requiring authors to declare their use of AI tools, should journals instead require them to specify their intellectual contribution to both the research and the manuscript? At the same time, publishers must devote much greater attention to fact-checking, source verification, and bibliographic validation. These additional responsibilities inevitably lengthen the review process and require greater investment in human expertise, financial resources, and technological support.
Another significant challenge concerns AI-detection systems themselves. Given the rapid evolution of large language models, no existing detection system can reliably identify AI-generated text with complete accuracy, even when multiple professional tools are used together. Moreover, we are increasingly encountering false positives, particularly in manuscripts by experienced scholars whose writing is naturally fluent and highly polished.
Open science and advances in machine translation introduce another layer of complexity. Today, authors from many countries can prepare manuscripts primarily from sources published in their native languages and submit them in English. In such cases, both plagiarism-detection software and AI-detection systems often perform poorly. For publishers, these submissions can become something of a “black box”, making thorough fact-checking and bibliographic verification extremely difficult. Taken together, these developments require a fundamental reassessment of editorial policy.
Many journals, including ours, now require authors to disclose the use of AI tools in manuscript preparation. We also subject every submission to multiple AI-detection systems before peer review. However, these measures alone are unlikely to be sufficient. The broader question remains: what additional mechanisms must scholarly publishers develop to ensure that academic publications continue to reflect genuine human scholarship and trustworthy scientific knowledge? That, I believe, is one of the central questions we will explore together in this seminar. Thank you.
Transforming Governance Through Artificial Intelligence
Wing Commander S. Sudhakaran (Retd)
Today I would like to discuss an aspect of artificial intelligence that is both fundamental and surprisingly underexplored. My presentation is titled “Transforming Governance Through Artificial Intelligence: Intelligence Beyond Automation”. Much of the current discourse on AI focuses on automation—making processes faster, cheaper, and more efficient. But I would like to ask a more fundamental question: Where does automation end and intelligence begin?
The answer has profound implications for governance and public policy, particularly regarding today’s seminar topic—how editors and publishers should govern the use of AI in scholarly publications. To answer that question, we must first clarify what we mean by the term “artificial intelligence.”
One of the greatest challenges today is that people often use the same term to refer to very different concepts. Broadly speaking, I see three groups of people engaged in discussions about AI. The first group comprises non-technical observers. These are intelligent people with no formal background in AI who have formed opinions because AI has become part of everyday conversation. For many of them, AI appears almost magical—a technology capable of performing tasks once considered uniquely human. Others view it as a potential threat, raising concerns about employment, ethics, and the future of society. Both perspectives are understandable, as they arise without direct technical engagement.
The second group comprises technically literate non-practitioners. These are individuals with engineering or scientific backgrounds who understand computing but have never built or worked extensively with AI systems. They often equate AI with highly sophisticated automation—a faster computer, a more capable calculator, or a more efficient digital assistant. While these capabilities are important, they primarily describe automation rather than intelligence.
The third group comprises AI practitioners—those who design, develop, and work directly with these systems. Their perspective often differs because they focus less on what AI automates and more on whether it actually performs cognitive acts.
This distinction is critical. The famous story of the blind men describing an elephant offers a fitting analogy. Each person experiences only one part of the elephant and therefore reaches a different conclusion about what it is. Today’s debate on AI often resembles that story. Different communities observe different aspects of the technology and naturally arrive at different definitions. Unfortunately, many public policies are largely based on the first two perspectives. We often begin by defining AI as automation and then extrapolate that definition to intelligence. As a result, policy often oscillates between excessive caution and excessive optimism. Both responses stem from attempts to regulate a phenomenon that has not yet been clearly defined.
That brings me to a more fundamental question. What do we actually mean by intelligence? For many years, I have argued that intelligence comprises three distinct components, which I call the Three Cs of Intelligence. The first is Comparison.
If I ask, “What is the capital of India?” and you answer “New Delhi,” you are retrieving stored knowledge. Someone taught you the answer; you retained it and recalled it when needed. This is intelligence based on comparison with stored memory. Even a trained sniffer dog identifying a familiar scent demonstrates this form of intelligence.
The second is computation. If I ask you to calculate the area of a circle, you first need the radius. You then apply a mathematical formula to obtain the answer. This is computational intelligence—processing information using established rules and procedures. Computers have always excelled at these two forms of intelligence. They retrieve information at extraordinary speed and perform calculations at scales impossible for human beings.
The third and most significant component is Cognition. Cognition encompasses understanding, judgement, creativity, contextual reasoning, and ultimately responsibility for knowledge. It is the ability not merely to retrieve or calculate information but to generate new understanding. Scientific breakthroughs rarely arise from comparison or computation alone. They stem from cognitive insight—the uniquely human capacity to formulate hypotheses, challenge assumptions, and generate new knowledge.
This distinction is essential because today’s generative AI systems remain overwhelmingly comparative and computational. They are extraordinarily capable within those domains, but they are not yet fully cognitive in the human sense. This is not simply a temporary technological limitation awaiting the next software update. It marks a fundamental distinction between automation and genuine cognition. However, the trajectory of development is unmistakable. The field is steadily moving from systems that compare and compute towards systems that increasingly simulate aspects of cognition. Whether that transition occurs next year or several years from now is difficult to predict, but the direction of travel is clear.
I describe this evolution as an intelligence maturity curve. In its early stages, machines primarily assisted human work through comparison and computation. Later, they augmented human capabilities by generating drafts, analysing information, and supporting decision-making. Today, we find ourselves at an important transition point.
Not long ago, humans produced every piece of scholarly work independently. More recently, authors began using digital tools to support their writing. Today, many researchers increasingly act as verifiers of machine-generated content—reviewing, refining, and validating material produced by AI systems. The next stage may see humans functioning primarily as overseers rather than active verifiers. Beyond that lies a future in which human involvement could diminish further as increasingly autonomous systems emerge.
Whether that future arrives soon or not, it underscores the need to clearly distinguish between automation and cognition. Referring to every stage simply as “AI” obscures profound differences in capability and responsibility. This distinction is particularly important in scholarly publishing. If we misunderstand the nature of these technologies, our editorial responses are likely to be ineffective.
For example, AI detection systems cannot reliably identify a category that is poorly defined. False positives are increasingly common, sometimes affecting experienced researchers whose writing is simply clear and sophisticated. Similarly, mandatory disclosure policies may unintentionally discourage openness if authors believe that declaring the legitimate use of AI tools will automatically prejudice editorial judgement.
Another challenge is what I call the “fog of technology”, borrowing from the military concept of the “fog of war”. Editors, reviewers, authors, and even developers often have limited visibility into how AI systems generate their outputs. We frequently cannot determine with certainty which ideas originate with the author, which emerge from the machine, and how the two interact. Consequently, I believe we are asking the wrong question.
Rather than asking, “Did a machine contribute to this manuscript?” we should ask, “Where does the author’s cognitive contribution reside?” That is a question editors can meaningfully evaluate. Authorship is not merely the production of text. It is the willingness to stand behind a claim, to defend its accuracy, and to accept responsibility for its consequences. Today, despite the remarkable capabilities of generative AI, responsibility still rests with the human author. The machine may assist with generating text, organising ideas, or retrieving information, but it does not assume responsibility for the truth of a scholarly argument. Responsibility requires understanding, intent, and accountability. These remain fundamentally human attributes.
For that reason, I believe editorial policy should focus less on detecting machine generated language and more on safeguarding the cognitive integrity of scholarship. The central question is not whether AI was used, but whether the human author remains intellectually responsible for the work. If we uphold that principle, our editorial policies will naturally evolve in the right direction.
If we lose sight of it, no amount of detection software or procedural regulation will adequately safeguard scholarly integrity. Let us therefore build our governance frameworks on first principles rather than on technological reflexes. Let us protect the cognitive core of scholarship while embracing technologies that genuinely enhance human research.
Thank you.
Questions and Discussion
Prof. Lilia Kamolova
Thank you very much for a highly thought-provoking presentation. I believe it offers a useful framework for understanding how editors should approach AI-assisted manuscripts. One idea that particularly resonated with me was your emphasis on cognition as the defining characteristic of genuine scholarship. I believe this is likely to become one of the central issues that editors and publishers will need to address over the next two or three years. My first question is this: given the rapid pace of technological development, do you think we should begin preparing now for what you described as the fourth stage of the intelligence maturity curve, rather than waiting until it arrives? My second question concerns young researchers. As editors, reviewers, and academic supervisors, we increasingly encounter scholars who routinely use AI tools in their research and writing. How should we approach these cases? If we recognise that these systems demonstrate sophisticated forms of intelligence, albeit not human cognition, what principles should guide our evaluation of their work?
Wing Commander S. Sudhakaran (Retd)
Thank you, Professor, for those important questions. Let me begin with your first question. The fourth stage marks a fundamental transition. Once that threshold is crossed, we are no longer dealing merely with systems that automate human tasks; we are entering an era in which machine intelligence could rival—and potentially surpass —human cognitive capabilities. From a cybersecurity perspective, this transition would be an inflexion point, leading to exponential growth in capability. It is closely associated with what is often described as the AGI (Artificial General Intelligence) singularity.
For several years, I have argued that this is the point we should be preparing for, as it fundamentally alters our assumptions about human oversight and control. Regardless of where one believes we are on that timeline, I think the most important lesson is to return to first principles. Rather than becoming preoccupied with labels or fashionable terminology, we should ask fundamental questions about what intelligence is, what cognition entails, and where responsibility ultimately lies. Technology evolves rapidly, but clear thinking remains our greatest advantage.
Turning to your second question about young researchers, I believe AI tools are now integral to the research environment. Trying to prevent their use altogether is neither practical nor, in my view, desirable. History offers many examples of technologies initially viewed with suspicion. Calculators were once prohibited in classrooms and examinations because they were believed to undermine learning. Scientific calculators capable of storing mathematical formulae were similarly restricted. Yet today we accept these tools because they allow people to devote more attention to higher-order reasoning rather than routine calculation. The same principle applies to AI.
The volume of scientific literature has grown to such an extent that no individual researcher can reasonably memorise or process it all. Used responsibly, AI can assist with searching, organising, summarising, and presenting information. These capabilities are valuable. The critical question, however, is whether the researcher continues to exercise independent judgement. The machine may generate text or retrieve information, but it cannot assume responsibility for the scientific claim. That responsibility must always remain with the human author.
As technologies continue to evolve—including developments such as brain-computer interfaces—we will increasingly encounter situations in which the line between human capability and technological augmentation blurs. These developments will undoubtedly raise difficult ethical and regulatory questions, and there are unlikely to be simple answers. What matters is that we establish sound principles now, rather than reacting after the technology has overtaken us.
Ultimately, this is not only an issue for individual researchers but also for nations. Countries that develop a deep understanding of these technologies and build indigenous capabilities will be far better placed than those that remain dependent on others. For countries such as India and Russia, technological sovereignty is therefore not merely an economic objective; it is a strategic necessity. That is why I believe our discussion today is so important. Before we formulate policies or regulations, we must first ensure we understand the technology itself. If our understanding is sound, our policies are far more likely to be effective. If it is flawed, no amount of regulation will compensate for that weakness.
Economic Cost Shifting of AI in Scholarly Publishing
Prof. Alexey Kuznetsov (INION RAN)
Much has been written and said about the benefits of artificial intelligence. We hear about improvements in productivity, efficiency, and accessibility. We also discuss its ethical, social, psychological, and even philosophical implications. However, one dimension remains surprisingly underexplored: the economic cost of AI. In my view, discussions of AI implementation often underestimate these costs, particularly in scholarly publishing and research. From an economic perspective, artificial intelligence should achieve one or both of two objectives.
First, it should save time. Second, it should reduce overall personnel and infrastructure costs. Yet this is often not what we observe in practice. Many organisations introduce sophisticated AI systems only to find they require additional technical specialists, new software licences, expanded digital infrastructure, and increased editorial oversight. The technology promises efficiency, but overall costs frequently rise rather than fall. Even more importantly, these costs are often shifted rather than eliminated. The author saves time, but the journal spends more time. The researcher becomes more productive, but the publisher incurs higher editorial costs. Ultimately, society pays for this redistribution of effort.
Drawing on my experience as Editor-in-Chief, I would like to highlight three areas where this cost shift is particularly evident in the social sciences and humanities. First, literature searches and reference verification. Artificial intelligence is undoubtedly a useful supplementary tool for literature searches. When used alongside traditional bibliographic methods, it can help identify publications that might otherwise be difficult to locate. For example, if we possess only an author’s surname—and it happens to be a common one—AI can often identify related publications much more efficiently than conventional searches. However, these benefits come with significant drawbacks.
Large language models frequently generate fabricated references or inaccurate citations. Even when the cited publication genuinely exists, AI-generated summaries may subtly misrepresent the author’s original argument. Such inaccuracies are often difficult to detect because they appear plausible. From an editorial perspective, verifying these references requires substantial additional work. In many cases, this effort is comparable to the time required for professional language editing, effectively doubling the editorial workload.
This is particularly significant because most scholarly journals operate on limited budgets and are not profit-making. Based on our internal estimates, the additional editorial effort devoted solely to verifying AI-generated references amounts to approximately US$1,000 per journal per year. That may appear modest in isolation, but across hundreds or thousands of journals it becomes a substantial financial burden.
Second, the increased burden on peer review. AI is undoubtedly useful for language editing, formatting, translation, and the preliminary review of large volumes of material. However, these same tools also make it much easier to produce manuscripts that appear professionally written yet offer little genuine scholarly contribution. Consequently, journals receive more submissions that satisfy formal editorial requirements but still require detailed expert evaluation.
The burden therefore shifts to peer reviewers. Most reviewers perform this work voluntarily. Their time is an important form of academic social capital. Every additional manuscript requiring review consumes scarce expert resources that could otherwise support genuinely original research. If we assign even a modest economic value to expert peer review, the cumulative costs become considerable. Using current Russian estimates for professional expert review, one additional review represents approximately US$50 of expert time. If AI-generated submissions create only one hundred additional reviews annually, the indirect cost exceeds US$5,000 per journal. Again, these costs are rarely included in discussions about the economic benefits of AI.
Third, the inappropriate use of AI for analysis and conclusions. In my opinion, this is the most serious issue. AI can assist with searching, organising, translating, and even drafting text. What it should not replace is scholarly analysis and intellectual judgement. Recently, I reviewed a manuscript submitted by a highly respected research institution. The institution itself enjoys an outstanding reputation built over decades by the work of distinguished scholars. However, the concluding section—including future scenarios and policy recommendations—had clearly been generated by a large language model. The recommendations concerned international political developments and strategic policy. This was not genuine scholarly analysis. It was the output of a language model presented as expert judgement.
For an editor, this raises profound concerns. Technical specialists increasingly acknowledge that AI-detection systems will become less reliable as these models continue to improve. If such material cannot be reliably identified, it risks entering the scholarly record and, eventually, informing public policy. In disciplines such as international relations or strategic studies, that possibility warrants serious attention. As Editor-in-Chief, rejecting such a manuscript is straightforward when the problem is obvious. The greater concern is how many similar cases remain undetected across the wider academic publishing ecosystem.
I have attempted to estimate the broader financial impact of these developments. A typical publicly funded research project in Russia may involve funding equivalent to approximately US$7,000 before culminating in a published paper. When we combine the costs of verifying AI-generated references, additional peer review, editorial oversight, and the occasional publication of low-quality AI-assisted research, the total additional cost can reasonably be expected to range from US$10,000 to US$20,000 per journal each year.
Across approximately one thousand journals in the social sciences and humanities, this represents an annual cost of between US$10 million and US$20 million. To put that figure in perspective, it is roughly equivalent to the annual funding required to operate two major research centres. In other words, the hidden costs of inappropriate AI use could consume resources that might otherwise support genuine scientific research.
My purpose today is not to argue against artificial intelligence. Used appropriately, AI is a valuable research assistant. However, before celebrating its economic benefits, we must also account for its hidden costs. These costs are often shifted from authors to journals, from researchers to reviewers, and ultimately from individual users to society. Only by measuring both the benefits and the costs can we develop balanced editorial policies and make informed decisions on the responsible use of AI in scholarly publishing.
Thank you
Questions and Discussion
Prof. Sunaina Singh (India Foundation):
What I find particularly valuable is that you have drawn our attention not only to the direct costs of AI but also to how those costs are being transferred across the scholarly publishing ecosystem. This is clearly a challenge that extends well beyond Russia or India and affects academic publishing worldwide. As AI tools such as ChatGPT, Gemini, Claude, and others become increasingly sophisticated, journals will inevitably require more editors and reviewers to evaluate submissions, verify sources, assess originality, and ensure compliance with ethical standards. At the same time, one concern continues to trouble me. As AI becomes more capable of generating fluent academic prose, are we gradually losing sight of the importance of critical thinking and independent intellectual inquiry? How do you think publishers should address these growing challenges while maintaining academic quality and economic sustainability?
Prof. Alexey Kuznetsov:
I believe one of the central challenges is the limited cooperation between the organisations developing AI technologies and the academic institutions that must address their consequences. From an editor’s perspective, some of the most difficult submissions come from researchers who work in artificial intelligence. Naturally, they make extensive use of these tools, and that is entirely legitimate. The difficulty arises when current detection systems flag AI involvement without allowing editors to distinguish between appropriate assistance and inappropriate substitution of scholarly work.
Even more concerning is the persistence of fabricated references and inaccurate citations generated by some AI systems. These errors are not always easy to identify, yet they consume considerable editorial time. This is precisely why closer collaboration between technology developers and the academic publishing community is so important. Editors understand the practical challenges of maintaining research integrity, while technology companies possess the expertise needed to improve these systems. Both communities should work together.
I was reminded of this at a recent conference in Tunisia. While paying for dinner, the waiter relied entirely on a calculator for a simple calculation and accidentally presented a bill almost twice the correct amount. Fortunately, I noticed the error immediately. The story is trivial, but it illustrates an important point. Technology should support human judgement—not replace it. If we become accustomed to accepting machine-generated outputs without exercising our own critical thinking, we risk losing an essential intellectual skill.
The same principle applies to artificial intelligence. AI should remain an assistant rather than a substitute for human reasoning. Unfortunately, there are no simple policy solutions. Governments understandably encourage technological innovation, as AI is increasingly a driver of economic growth. At the same time, society must ensure that innovation remains aligned with the broader public interest.
For me, the answer ultimately lies in education. We must teach researchers, editors, and students not only how to use AI tools, but also how to question and verify their outputs, and to remain intellectually responsible for the conclusions they present. Technology will continue to advance. Our responsibility is to ensure that human judgement advances alongside it.
AI’s Potential vs. Human Insight: Ethical Anxiety for Scholarship and Originality
Prof. Sunaina Singh (India Foundation)
After listening to the excellent presentations this evening, I find myself with more questions than answers. Rather than discussing the technical aspects of artificial intelligence, I would like to reflect on three interconnected issues: AI and human insight, the ethical challenges it poses for scholarship, and the meaning of originality in an age of intelligent machines. These questions have occupied my thoughts for some time, particularly because I believe the humanities and the social sciences are uniquely positioned to lead the conversation on the responsible and ethical development of AI. Allow me to begin with a brief personal perspective.
My observations are shaped by my experience of leading two universities—one national and the other international—and by serving on a number of academic and regulatory bodies responsible for policy formulation. Throughout my career, I have been involved in designing institutional policies on research, teaching, governance, and academic standards. One lesson has become increasingly clear to me. Policies cannot remain static.
In an age of rapid technological change, governance itself must become agile. Policies must be reviewed regularly, adapted to emerging realities, and refined as technology evolves. This is particularly true of artificial intelligence, where the pace of innovation frequently outstrips regulation. We are, in my view, living through one of the most transformative periods in human history. Artificial intelligence is reshaping economies, institutions, education, research, and even the way knowledge itself is created and disseminated.
Yet we also live in a world marked by conflict, uncertainty, inequality, and profound societal challenges. This raises an important question. Can AI help us address these challenges? My answer is both yes and no. AI can certainly process enormous volumes of information, identify patterns, analyse data, and present possible alternatives with remarkable speed. But information is not the same as wisdom. Data is not the same as judgement. Knowledge is not the same as understanding.
These distinctions are especially important in scholarship. Artificial intelligence can synthesise existing knowledge, but it cannot exercise moral judgement or ethical reasoning independently. It cannot determine what ought to be done simply because it can explain what has been done before. That remains the responsibility of human beings. This is precisely where the humanities and social sciences play a vital role.
Unlike many other disciplines, these fields are fundamentally concerned with human experience—values, ethics, culture, history, identity, justice, and the complexities of society. They encourage us to ask difficult questions rather than merely produce efficient answers. They help us understand not only how societies function but also how they ought to evolve. That is why I believe these disciplines must play a leading role in shaping AI governance.
Many institutions have already developed policies governing the use of artificial intelligence in teaching and research. This is a welcome start. However, policy cannot be treated as a one-off exercise. As AI systems become more sophisticated, our ethical frameworks must evolve alongside them. Responsible governance requires continuous review, learning, and adaptation.
Another concern that increasingly occupies my mind is the risk of intellectual passivity. Today, answers are available almost instantaneously. Whether we consult ChatGPT, Gemini, Claude, or another AI system, information is at our fingertips within seconds. This convenience is undoubtedly valuable, but it also raises an important question. If machines increasingly provide answers, are we becoming less inclined to ask questions? Are we gradually losing the habit of reflection, critical inquiry, and independent reasoning? In other words, are we becoming intellectually passive?
I do not yet have a definitive answer, but I believe it is a question that educators, researchers, and policymakers must take seriously. Knowledge has never been merely the accumulation of information. True education is about cultivating curiosity, developing critical thinking, exercising sound judgement, and engaging thoughtfully with complexity. These are profoundly human capacities. Artificial intelligence can certainly support them. It can organise information more efficiently than any individual researcher, help us discover connections that might otherwise remain hidden, and serve as a valuable research assistant. But it cannot replace the intellectual and ethical responsibility at the heart of scholarship.
This distinction is particularly important in the humanities and social sciences. In many STEM disciplines, AI already offers remarkable capabilities for modelling, simulation, optimisation, and data analysis. The humanities and social sciences, however, engage with human emotions, values, cultures, institutions, and societies. These are domains where context, interpretation, empathy, and ethical judgement remain indispensable. For that reason, I believe we must approach AI differently in these disciplines. Rather than asking whether AI can replace human scholarship, we should ask how it can strengthen human scholarship while preserving the qualities that make it uniquely human. Ultimately, I believe the future lies not in choosing between artificial and human intelligence, but in creating a thoughtful partnership between the two.
Artificial intelligence can expand access to knowledge, preserve cultural traditions, organise vast bodies of information, and enhance research productivity. Human beings must continue to provide judgement, wisdom, ethical reasoning, and intellectual leadership. Perhaps that is the challenge before all of us. As I said at the beginning, I leave this discussion with more questions than answers. But perhaps that is not a weakness. After all, scholarship has always advanced by asking better questions before finding better answers. I hope our discussions today will contribute to that ongoing journey.
Thank you.
Stigmatised Artificial Intelligence Use: AI-Friendly Journal Policies vs. Quasi Sensorial Editorial Practices
Sergey Gonov (INION RAN)
Before turning to my own presentation, I would like to briefly comment on a point raised by Professor Kuznetsov about fabricated quotations and references generated by AI. The problem is not confined to the original fabrication. Its real danger lies in what I would call the long citation trail. An author unknowingly incorporates a fabricated quotation generated by a chatbot into a published paper. Another scholar subsequently cites that publication in good faith. The fabricated quotation is then repeated across multiple publications until, perhaps years later, an improved AI-detection system identifies its origin. At that point, it is not only the original paper that becomes questionable, but an entire chain of subsequent scholarship built upon it. In other words, fabricated AI content has the potential to contaminate the scholarly record long after it first appears.
That observation brings me to my own topic. Today I would like to examine not the technology itself, but the way academic institutions are beginning to respond to it. My argument is straightforward. We are gradually developing policies that encourage transparency about AI use, while simultaneously creating editorial cultures that penalise it. This contradiction warrants careful attention. Allow me to illustrate it with several real academic cases.
The first concerns a master’s dissertation. For an entire year, the supervisor had worked closely with the student. He had observed the research process, discussed the literature, reviewed successive drafts, and witnessed the project’s intellectual development. Naturally, his evaluation was highly positive. A second examiner, however, saw only the finished manuscript. Although many of the comments concerned technical details, one underlying suspicion coloured the entire review: the possibility that artificial intelligence had been used.
The dissertation itself had not changed. Only the mode of reading had changed. One evaluator interpreted the work in the context of a year of academic supervision, while the other approached it primarily as a potentially AI-generated text. This illustrates two fundamentally different editorial mindsets. In one, AI is regarded as a legitimate research tool whose use can be regulated. In the other, AI remains primarily a symbol of academic dishonesty.
A second example comes from Kazan Federal University. A graduate student was expelled after an AI detector flagged her thesis as containing AI-generated content. The Supreme Court of Tatarstan later overturned the decision. The Court’s reasoning is particularly significant. It held that the output of an AI-detection system should be regarded as an indicator—not as conclusive evidence. That appears entirely reasonable. Yet it immediately raises another question. How are such indicators interpreted in practice?
Many instructors continue to rely on freely available online detection tools and treat their results as definitive proof of misconduct. Even institutional systems require careful interpretation. In Russia, for example, the national plagiarism-detection platform includes an AI detection module. The detailed report is appropriately cautious, referring to “suspected AI-generated fragments” or indicating that certain passages are “likely” to have been generated by AI. However, the summary page presents numerical indicators in a far more categorical manner. Readers see boxes that report percentages of originality, quotations, and AI-generated content. Although the underlying algorithm expresses probability, the visual presentation often resembles an official verdict. A statistical estimate is transformed into a social label. Probability becomes stigma.
Yet the opposite problem also exists. In one editorial case, an AI detector reported no suspicious content, so the manuscript appeared perfectly acceptable. However, during manual review, a referee noticed that one sentence had been copied directly from the chatbot’s interface: “Would you like me to develop this topic further and provide additional examples?”
The detector had entirely missed the problem. The manuscript was rejected—not because AI had been used—but because the author had shown no meaningful editorial engagement with the generated text. This illustrates an important point. Detection systems can fail in both directions. They may incorrectly stigmatise legitimate scholarship, and they may also fail to identify careless or inappropriate uses of AI. Naturally, this leads us to rely on expert judgement.
But another difficulty arises. The Supreme Court rightly observed that final decisions should involve expert evaluation. But who exactly qualifies as an expert in AI generated writing? Most academics are well qualified to evaluate argumentation, methodology, sources, evidence, and scholarly style. Determining the technical origin of a piece of writing is a different matter entirely.
Allow me to offer one final example. Our journal received a manuscript from a senior scholar I knew personally. The article was stylistically weak, loosely organised, and excessively literary for an academic publication. A reviewer immediately concluded that these characteristics indicated AI use. Ironically, my own experience has generally been the opposite. Most AI-generated texts are highly structured, formally organised, and often excessively coherent. The review therefore reflected not expertise in AI but assumptions about it. This distinction is important.
Expert judgement is not always expert knowledge. Sometimes it is simply personal belief. What conclusions can we draw? Today, many journals require authors to disclose their use of artificial intelligence through formal AI declarations. I strongly support transparency. However, transparency alone is insufficient. We also need ethical standards to govern how disclosures are interpreted. In other words, policies should regulate not only authors but also reviewers and editors. Otherwise, disclosure itself becomes risky.
If authors believe that admitting to using AI automatically casts their work in suspicion, many will simply choose not to disclose it. The unintended consequence is that a policy designed to encourage honesty may instead discourage it. This is what I call a quasi censorial effect. The problem is therefore not only the ethics of AI use but also the ethics of AI suspicion. Disclosure should be a neutral statement of methodology—not an implicit admission of academic misconduct.
Only when transparency carries no automatic stigma will authors have genuine incentives to disclose AI’s role in their research. That, I believe, is one of the next major challenges facing scholarly publishing.
Questions and Discussion
Shrishti Pukhrem (India Foundation):
I found your discussion of the stigma surrounding AI use particularly thought provoking. As researchers working in international relations and the social sciences, we are witnessing the rapid integration of AI into academic practice, making these questions increasingly relevant. Your comments on disclosure policies merit special attention. Many leading academic publishers now require authors to include explicit AI-use statements, specifying whether AI tools were used, which tools were employed, and for what purposes. At the same time, these policies make it clear that responsibility for the accuracy, originality, and integrity of the work remains entirely with the human author. This approach has been adopted by several major publishers, including Springer Nature, Wiley, and Sage.
However, your presentation prompted me to think beyond the responsibilities of authors and towards those of peer reviewers. What happens when reviewers themselves use AI during the review process? This raises a number of important questions. If a reviewer uploads an unpublished manuscript to a public AI system to draft a review, are confidentiality obligations compromised? Manuscripts submitted for peer review are confidential documents containing unpublished intellectual property. Their use on external AI platforms raises concerns about privacy, data security, and ownership.
A second issue concerns reviewers’ use of AI-detection systems. How reliable are these systems, and to what extent should their outputs inform editorial judgement? These questions suggest that editorial policies should address not only authors’ responsibilities but also reviewers’ ethical responsibilities. I would also like to connect this discussion to a point Professor Sunaina Singh raised earlier about equity and access. Access to advanced AI tools remains highly uneven. In India, many researchers at smaller institutions or in economically disadvantaged regions lack access to premium AI platforms. I imagine similar disparities may exist in other countries, including Russia. If sophisticated AI assistance gradually becomes an implicit expectation in scholarly publishing, it could unintentionally widen existing inequalities by favouring researchers who can afford advanced tools over those who cannot. Responsible AI governance, therefore, must also take equity and accessibility into account.
Sergey Gonov:
I fully agree that the role of reviewers warrants much greater attention. The use of AI by reviewers is already a significant concern, not only in scholarly publishing but also in university examination processes. We are increasingly encountering referee reports and thesis evaluations that appear to have been drafted, at least in part, with generative AI systems. This reinforces the central argument of my presentation.
If journals set rules for authors’ use of AI, they should also establish comparable ethical standards for reviewers and editors. The integrity of peer review depends on transparency, confidentiality, and independent intellectual judgement. These principles should apply equally to all participants in the publication process.
Prof. Alexey Kuznetsov:
I would like to add an observation on Sergey’s important point about stigmatisation. As Chair of examination committees at the Higher School of Economics, I encounter similar challenges when assessing graduate dissertations. Many students now submit formal declarations acknowledging the use of AI. Almost invariably, they state that AI was used only for language editing or translation, particularly in programmes where theses are written in English. However, examination committees frequently find that AI has played a much more substantial role in generating or developing the text itself. This creates a difficult situation. Supervisors are often reluctant to confront these issues openly because they fear that acknowledging extensive AI use may reflect poorly on their supervision or on the academic standards of their programme. As a result, concerns are sometimes overlooked until the dissertation reaches the examination
committee.
The committee is then placed in the difficult position of determining the extent of AI involvement without having observed the student’s research process. In this sense, stigmatisation has unintended consequences. Rather than encouraging honest disclosure and constructive discussion, it can create incentives for both students and supervisors to minimise or conceal the extent of AI use. For that reason, I believe Sergey’s concept of the “stigma of AI” is highly relevant not only to scholarly publishing but also to higher education more broadly. Unless we create an environment in which the use of AI can be discussed openly and responsibly, transparency itself may become the first casualty.
Formulating Editorial Standards and Addressing Structural Injustices in AI Governance
Prof. Sushma Yadav
Thank you. This subject is particularly close to me because it brings together several roles I have had the privilege of undertaking over the years—as Vice-Chancellor, former Pro-Vice-Chancellor of the Indira Gandhi National Open University, member of the University Grants Commission, member of editorial boards of scholarly journals, reviewer, and Chairperson of the UGC Empowered Committee responsible for the UGC-CARE journal list. These experiences have reinforced an important lesson: good scholarship depends on good governance, and good governance depends on policies that evolve with changing realities.
Artificial intelligence is now compelling us to rethink those realities. For the social sciences, this discussion is particularly significant. These disciplines are founded on the understanding that human experience, social context, ethical judgement, and interpretive reasoning are not merely inputs to research—they are integral to the creation of knowledge itself. This raises an important question. What does it mean for disciplines grounded in human understanding to accept scholarly texts produced, wholly or partly, by artificial intelligence?
Before we formulate editorial policies, we must first determine precisely what we are trying to govern. Not every use of AI in research is equivalent. Treating all AI applications as though they pose the same ethical challenge inevitably produces policies that are either too restrictive or impossible to implement. Several of the earlier speakers have made this point from different perspectives, and I believe it deserves emphasis. There are many areas in which AI can legitimately support scholarly work. Language editing, grammar correction, proofreading, formatting, translation, and improving clarity of expression are all appropriate uses of AI, provided they are transparently disclosed. These applications assist the researcher without replacing the researcher’s intellectual contribution.
However, certain dimensions of scholarship cannot be delegated to AI. The first is positionality and reflexivity. Researchers inevitably bring their own experiences, perspectives, ethical commitments, and relationships to their field of study. These are deeply human dimensions of scholarship that no AI system can genuinely reproduce.
The second is contextual interpretation. Understanding social realities requires historical awareness, cultural sensitivity, and lived experience. Language models can identify patterns, but they cannot fully grasp the social meaning that gives them significance.
The third is normative judgement. Much social science research requires scholars to make and defend value judgements. Deciding what ought to be done is fundamentally different from describing what has already been done.
Finally, there is ethical accountability. Researchers remain accountable to their participants, their institutions, and the wider scholarly community. Artificial intelligence cannot assume that responsibility.
These distinctions point to an important principle for editorial governance. Editorial policies should distinguish between AI used as a research support tool and AI used as a substitute for human intellectual contribution. The former is compatible with academic integrity when properly disclosed. The latter fundamentally misrepresents the nature of scholarly work, regardless of the quality of the resulting text.
Effective editorial governance therefore requires more than broad statements of principle. It requires a practical framework. In my view, every journal developing an AI policy should address four fundamental questions. First, what forms of AI use must be disclosed? Clear disclosure should be the minimum standard for scholarly publishing.
Second, what forms of AI use remain unacceptable even with disclosure? Some responsibilities—analysis, interpretation, scholarly judgement, and original argumentation—cannot be delegated to AI.
Third, how should journals verify compliance? This is perhaps the most difficult question. As the previous presentation reminded us, even reviewers may now rely on AI when evaluating manuscripts. How should journals respond? Current AI-detection systems remain probabilistic rather than definitive. They generate false positives and frequently disadvantage multilingual scholars whose writing patterns differ from those of native English speakers. Detection tools should therefore assist editorial judgement, not replace it. Ultimately, the most effective safeguards are cultural rather than technological. Journals should encourage honest disclosure, publish clear AI policies, review those policies regularly, and apply them consistently and proportionately.
Fourth, how should disclosures be interpreted by editors and reviewers? Disclosure should trigger neither automatic acceptance nor automatic rejection. Instead, it should prompt careful evaluation of the genuinely human intellectual contribution in the manuscript. This requires a new editorial competency. Many experienced editors and reviewers did not begin their careers in the age of generative AI. Journals must therefore invest in training editors and reviewers so they can confidently evaluate AI assisted scholarship.
Another issue that warrants far greater international attention is equity. Professor Sunaina Singh and Dr Shrishti Pukhrem have already highlighted this important concern. Much of the current global discussion is shaped by large, well-resourced, English-language publishing organisations. Yet the realities of the Global South are often very different. Researchers working in their second or third language are more likely to use AI for language support. Ironically, they are also more likely to be disadvantaged by AI-detection systems that misidentify non-native writing patterns as machine-generated. This represents a structural inequity that editorial policies must address.
The same applies to early-career researchers, first-generation scholars, and resource constrained institutions. Volunteer-run journals in developing countries cannot simply absorb the additional financial and administrative costs of increasingly complex AI compliance systems.
India provides an important example. Our research community publishes in more than twenty-two officially recognised languages. AI-assisted translation and language support have enormous potential to broaden participation in scholarship and strengthen research in regional languages. Editorial policies that treat all AI-assisted language support with suspicion risk undermining that opportunity.
Perhaps this is an area where India and Russia, as two major centres of scholarship beyond the traditional Anglophone publishing system, can make an important contribution. Could we collaborate to develop a shared framework for equitable AI governance in scholarly publishing? A joint statement recognising AI-assisted language support as a legitimate accessibility measure—rather than an academic integrity violation—would make a valuable contribution to international policy discussions.
In higher education, I believe three priorities merit particular attention. First, AI literacy should become integral to doctoral research training. Students need not only technical skills but also ethical understanding and sound scholarly judgement. Second, universities should establish clear institutional frameworks for AI-assisted research, distinguishing appropriate academic support from unacceptable substitution. Third, our systems of academic evaluation must continue to reward originality, critical thinking, intellectual depth, and societal relevance rather than simply counting publications.
Ultimately, the future of responsible AI governance will depend less on technology than on academic culture. Policies set minimum standards. Culture determines whether those standards are genuinely upheld. India’s National Education Policy 2020 rightly emphasises critical thinking, ethical reasoning, multidisciplinary learning, and research excellence. These principles become even more significant in the age of artificial intelligence. The challenge for higher education is not simply to adopt new technologies, but to ensure that innovation strengthens rather than weakens academic integrity and intellectual autonomy.
Allow me to conclude by proposing five principles that could form the basis of a shared editorial framework.
• First, disclosure should be the cornerstone. Any significant use of AI in preparing a scholarly manuscript should be disclosed.
• Second, differentiation should inform policy. Editorial standards must distinguish between AI that supports research and AI that substitutes for genuine intellectual contribution.
• Third, equity should be the criterion. Every policy should be assessed for its impact on multilingual scholars, early-career researchers, Global South institutions, and resource-constrained journals.
• Fourth, culture should take precedence over technology. Sustainable AI governance depends far more on professional ethics and scholarly norms than on increasingly sophisticated detection software.
• Finally, collaboration should be our guiding principle. India and Russia share long academic traditions and many common challenges. Working together through international organisations, learned societies, and editorial networks will enable us to contribute meaningfully to the development of fair, globally relevant editorial standards.
Ultimately, this discussion is not merely about artificial intelligence. It is about what we value in scholarship. We value rigour in producing knowledge. We value integrity in representing how that knowledge was created. And we value inclusion to ensure that scholarship remains accessible to the widest possible community of researchers.
Artificial intelligence can assist scholarship by accelerating research and improving communication. But it cannot replace human judgement, ethical reasoning, or scholarly responsibility. Our editorial policies should always reflect that fundamental truth.
Reconceptualisation of Scientific Information and Book Culture in the Age of Artificial Intelligence
Prof. Igor Buryanov (INION RAN)
Good afternoon. In this report, I will briefly outline my considerations on the place of the scientific book in the new world of artificial intelligence. Building upon the insightful observations of Professors Yadav and Kamolova, I would like to examine the relationship between artificial intelligence and the traditional infrastructure of knowledge, particularly books, libraries, and scholarly publishing.
Artificial intelligence is often presented as if it has access to the entirety of human knowledge. In reality, this assumption is misleading. Modern AI systems are built on vast collections of digitised books, journals, and other textual resources. The large scale digitisation of scholarly literature over the past several decades has undoubtedly transformed global knowledge production and provided much of the foundation for contemporary language models. Yet an important misconception persists. Many people assume that nearly all scholarly literature has already been digitised and is therefore available to AI systems. This is far from the truth. A substantial proportion of the world’s scholarly literature remains undigitised. Even in my doctoral research over the past year, I found that only about one-third of the sources I required were available online. The remaining material had to be located through physical libraries, archives, or printed collections. This experience illustrates an important reality. Artificial intelligence cannot analyse information that has never been digitised.
However, digitisation alone does not guarantee accessibility. Even when books and journals are available in digital form, they often remain inaccessible due to copyright restrictions, subscription barriers, institutional licences, or classified collections. Many valuable academic publications can be consulted only through university libraries, purchased individually, or obtained through specialised archival collections. Consequently, a significant body of knowledge is digitally preserved yet effectively invisible to most AI systems. This limitation becomes even more significant as the global information environment grows increasingly fragmented.
The fragmentation of the global internet is likely to inexorably, in the near future, lead to the fragmentation of knowledge itself, isolating academic disciplines and even entire nations from global scientific discourse. The evolution of AI largely mirrors these processes. Global data storage systems are on the rise, alongside local AI models accessible only within one or several countries and sealed off from others. Access to AI has already become a target of sanctions and restrictions imposed on Russia; for instance, Google has barred Russians from using Gemini.
Under these circumstances, the ability of AI models to draw on a diverse range of sources becomes critically important. Over time, an increasing share of specialised literature will be used only by isolated local AI models, and the interpretation of research data will be feasible only when a physical or digital copy of a publication is held by a specific individual or institution. This, in turn, compels us to rethink the role of print literature, particularly those books that, for one reason or another, have never been uploaded to the internet. Their number remains vast, and this includes a great deal of newly published material. A considerable portion of recent books and articles never enters the public domain, whether because of varying levels of classification or the detection of restricted content in their text.
Today, local models are becoming increasingly widespread, deployed at multiple levels, from states down to institutions. In the coming years, major libraries will not only build their own proprietary databases but will also develop local models to facilitate information discovery. A defining feature of these models will be their ability to mine content from publications that are often held in only one country, and sometimes in just one library. This will make it possible to access information contained in books whose online publication remains impossible due to legislative restrictions and copyright protections.
Thus, the fragmentation of the global information space simultaneously gives rise to the fragmentation of AI models and to the emergence of national AI services shaped by state information and ideological policies, as well as by the specific ways data are disseminated both online and in traditional print. Taken together, these factors already raise profound questions about AI’s capacity to adequately support scientific research, let alone to replace humans in compiling comprehensive and up-to-date literature reviews, peer reviewing, or abstracting. In this new stage of the internet’s evolution, marked by the advent of AI, book culture does not disappear; rather, it takes on a new quality and a new role, serving as a vital alternative to the all-encompassing pressure and influence of artificial intelligence in the advancement of science and technology.
Thank you for your attention.
Closing Remarks
Maj. Gen. Dhruv C. Katoch, Director, India Foundation
Thank you all for your insightful and thought-provoking presentations. Today’s discussions have highlighted not only the immense potential of artificial intelligence but also the ethical, editorial, economic, and institutional challenges that accompany its growing use in academic research and publishing. Many of the concerns raised resonate strongly with my experience as Editor of the India Foundation Journal.
Our deliberations have made it clear that no single institution or country can address these issues in isolation. They require sustained dialogue, collaborative research, and the sharing of best practices across academic communities. As a collaborative initiative, we could prepare a joint statement, accompanied by a set of practical recommendations that could serve as a policy paper on the responsible and ethical use of artificial intelligence in scholarly publishing.
Prof. Alexey Kuznetsov
Thank you very much. I agree that today’s seminar should be seen as the start of a longer conversation rather than a one-off event. Our discussions have highlighted both the complexity of these issues and the value of bringing together perspectives from different disciplines and institutional backgrounds. I believe we now have an excellent foundation for further collaborative work. I am confident that continued cooperation between our institutions will make a meaningful contribution to the international discourse on the ethical and responsible use of artificial intelligence in scholarly communication.
Thank you all once again for an excellent and highly productive discussion.
