How AI is shaking up scientific publishing

  how-ai-is-shaking-up-scientific-publishing.html
  Science and Technology on Sun, Aug 24th, 2025 by Le Monde.fr

Artificial Intelligence Is Redefining Scientific Publishing: What the Latest Report Says

The scientific community is already accustomed to the power of artificial intelligence (AI) in data analysis, modelling and hypothesis generation, but the article “How AI is shaking up scientific publishing” (Le Monde Science, 12 July 2025) shows that the same technology is beginning to transform the very act of publishing. The piece traces a rapidly evolving landscape in which AI tools are both a boon and a source of concern for authors, reviewers, and publishers alike.

1. AI‑generated manuscripts – the new “ghost writers”

One of the most startling trends the article highlights is the rise of AI‑driven text generation. Researchers now routinely use language models such as GPT‑4 or Claude to draft sections of their papers, from introductions to discussion points. A 2024 study cited in the article found that 35 % of submissions to an open‑access journal contained at least 10 % of their text generated by an AI. The result is a “ghost writing” problem: when an algorithm composes a paragraph, who owns the intellectual property? Several journals have begun to ask authors to disclose any AI assistance, and the International Committee of Medical Journal Editors (ICMJE) has updated its authorship guidelines to require that contributions from non‑human entities be explicitly acknowledged.

2. AI in peer review – a double‑edged sword

The review process itself is being reshaped. AI algorithms now scan manuscripts for logical inconsistencies, statistical errors and plagiarism with a speed that would take a human reviewer days. One example the article mentions is the “ScholarAI” tool adopted by Springer Nature, which flags suspicious language patterns and automatically cross‑checks references against databases such as PubMed and arXiv. While the speed gains are undeniable, reviewers worry that over‑reliance on automated checks could stifle the nuanced judgment that human experts bring. In response, the American Association for the Advancement of Science (AAAS) has announced a pilot program that pairs AI flagging with a small panel of human reviewers to verify the AI’s suggestions.

3. The role of publishers and policy makers

The article points to a wave of new policies that aim to manage AI’s dual role as a facilitator and a threat. Nature, in 2023, issued a formal statement declaring that “AI‑generated text will not be credited as authorship unless the AI is explicitly acknowledged as a tool.” Elsevier, on the other hand, has introduced a mandatory “AI usage statement” for all submissions, modeled on the European Union’s forthcoming Digital Services Act, which will require authors to declare the extent of AI contribution. The European Commission’s Digital Ethics Guidelines for Trustworthy AI include a clause that “scientific publications must be transparent about any non‑human assistance.”

4. AI as a democratizing force

Not all changes are seen as negative. The article presents several case studies in which AI has helped researchers in resource‑constrained environments. For instance, the Indian Institute of Science used a multilingual AI summarizer to translate complex experimental procedures into regional languages, thereby widening participation in international collaborations. Similarly, a group of African biomedical researchers used an AI‑driven literature mining tool to identify gaps in malaria vaccine research, a feat that would have taken them months to accomplish manually.

5. Concerns over bias, hallucination and misinformation

AI’s propensity for “hallucination”—producing plausible but incorrect information—is a central worry. The Le Monde piece cites a 2024 meta‑analysis that found 18 % of AI‑generated abstracts contained at least one factual error. Because these tools rely on statistical patterns rather than semantic truth, the risk of propagating misinformation is non‑trivial. Publishers are responding by integrating AI‑driven fact‑checking modules that cross‑reference claims against curated databases. The article also notes that certain AI models exhibit demographic biases in language usage, potentially skewing the tone or emphasis of a manuscript. This has prompted a number of journals to adopt “bias‑audit” checks for AI‑generated content.

6. The future: a new set of norms

The article concludes by highlighting how the scientific publishing ecosystem is in the midst of a cultural shift. Many editors now view AI as a “standard instrument” similar to reference managers or statistical software. However, the article emphasizes that “norms will lag behind technology.” Calls for a global consensus on AI disclosure, authorship, and editorial oversight are gaining momentum. A notable development is the launch of the AI‑in‑Science Working Group by the International Council of Scientific Unions (ICSU), which aims to produce a set of guidelines that balance innovation with integrity.

In sum, the Le Monde article paints a complex picture. AI is accelerating the pace of publishing, improving quality control, and opening doors for under‑resourced researchers. At the same time, it threatens to blur authorship lines, introduce biases, and compromise the trustworthiness of the literature. As publishers, researchers, and policy makers grapple with these challenges, the field of scientific publishing is likely to see a series of new standards and regulatory frameworks in the years ahead—an evolution that will shape the very fabric of scholarly communication.