Scientific publications: a fatal level of overproduction?
They are variously labelled too numerous, fraudulent or written by generative artificial intelligence: criticism of scientific publications is ramping up in a current context of overproduction. But does this mean we should give up on them? This was the subject of debate at the CNRS's seventh Open Science Day.
Something has been wrong in the world of scientific publications for many years now and the process is speeding up. In 2025, there will be nearly five million unpublished articles, compared to just one million in the 1980s. There has been a particularly sharp increase in the last decade with the main bibliographic databases growing by nearly 50% between 2016 and 2020. This phenomenon is unrelated to the +16% increase in the number of scientists worldwide in the same period1 and this level of disproportion is a significant burden on scientists who carry out peer-reviewing activities. This partly explains the increased number of fraudulent – and therefore retracted – articles which amounted to 10,000 articles in 2023 from an annual total of 3 million new publications.
Alain Schuhl, the CNRS's Deputy CEO for Science (DGDS), puts his finger on it thus: "The emergence of generative artificial intelligence means we're in a system that can't have the same standards as before because there are no longer any authors, reviewers or readers". The "announced demise of scientific publications" was discussed in detail at the CNRS's seventh Open Science Day on November 25th of this year.
Part of the day was given over to understanding the mechanics of this global crisis. Several speakers highlighted the role of 'paper mills' which are now driven by generative artificial intelligence but what proportion of scientific articles do these paper mills actually represent? The lowest estimates are between 1% and 3% but they illustrate a general phenomenon as described by Lionel Maurel, the deputy director for scientific publishing at the CNRS's Open Research Data Department (DDOR)2 . Mr Maurel explains that "between AI-generated articles and citation purchases, we're reaching the end of a cycle and a loss of meaning. These were previously marginal problems but they've been amplified by the author-pays model. This was an easy solution rather than representing structural changes within the scientific publishing industry and it has served to exacerbate pre-existing problems".
However, when it comes to scientific publications, paper mills and other fraudulent practices are only the tip of the iceberg. The phenomenal growth of scientific publications derives primarily from the organisation of research itself according to Vincent Larivière, a library science specialist from the University of Montreal1 . "The scientific community publishes too much. Also, AI is not going to solve the issue of peer review which stems from an overproduction of scientific articles. It's not a technical problem, it's a political issue".
Overproduction and platform-based research
What is at the root of this political problem? The fact that the assessment of researchers is still often linked to their volume of publications. "We publish more because we're actually encouraged to on an individual basis", sums up Vincent Larivière and sometimes this can reach border on the absurd. Didier Torny, a CNRS publication economics researcher2 , suggests that the aim of many authors is "not to be read rather than to be read" because their articles are primarily aimed to fit with bibliometric algorithms to artificially inflate their own CVs. The pressure to publish is also affecting scientific publishers. "For economic reasons, they now accept a greater number of articles for publication", explains Denis Bourguet, the co-founder and co-director of Peer Community In, a peer-reviewed preprint recommendation service.
This has now led to bibliometrics and economics actually becoming inseparable as Yves Citton3 pointed out in his 'carte blanche' lecture. This professor of literature and media draws a parallel between the music industry and the scientific publishing industry, remarking that "in 2023, 57% of music revenue came from streaming... and therefore depends on a platform that can potentially be bought and reoriented for ideological purposes, unlike physical formats such as CDs". The platform model is less well known than its counterparts in the music and audiovisual industries but has now also conquered the world of scientific publishing.
- 1Tenured professor at the University of Montreal's School of Library and Information Science and holder of the UNESCO Chair on Open Science.
- 2CNRS research professor in sociology and scientific delegate with the DDOR.
- 3Professor of Literature and Media at Paris-VIII University and member of the Institut universitaire de France.
This has not taken place without knock-on consequences for editorial production itself. The Information and Communication Science researcher Chérifa Boukacem-Zeghmouri1 goes so far as to talk about "platformised research", which has shifted the writing of scientific articles from an "editorial model to a flow model in which cost, speed and technical ease take precedence, like in radio and television". And yet, as this Lyon-based researcher puts it, "this media model is incompatible with science". In this context of systemic criticism of platform-based research that depends on commercial bibliometric databases, the CNRS recently announced via its DGDS that it will cut access in early 2026 to one of the most important of these databases, Clarivate Analytics' Web of Science. "The subscription to the Web of Science costs €1.4 million per year, which will be spent in another way – to support open science", explains Alain Schuhl. The CNRS is now encouraging its researchers to use open databases like OpenAlex which offers enhanced visibility for non-English-language journals along with a greater number of journals than the Web of Science. This is a crucial step in the organisation's open science policy and follows on from the decision to unsubscribe from Elsevier's Scopus bibliometric database on January 1st 2024.
Decorrelating the assessment of publications
So does it make sense to continue publishing scientific articles in these circumstances? Two publishers were invited to take part in a round table discussion where they defended the value of scientific publications as long as technical modifications are implemented. Olivier Dumon, Chief Product Officer with one of the world's leading publishers, Elsevier, announced his company sees scientific integrity as a real issue. Indeed, Elsevier has invested $20 million in expanding its integrity team from two people in 2023 to 120 today. With a team like this, "Elsevier ensures that the rejection rate for articles submitted for publication remains at around 80%", which itself is a guarantee of the quality of the papers that Elsevier publishes. Regarding the pressure exerted by the overproduction of articles on reviewers, Olivier Dumon points out that "an article is rejected 2.6 times before a journal accepts it" and praises the creation of a tool based on artificial intelligence that is "capable of identifying relevant peer reviewers rather than relying on publishers' personal networks".
- 1At Claude Bernard Lyon 1 University.
Emmanuelle Jannès-Ober, the representative at the conference of the association of French public publishers1 also looks at the issue through the prism of peer reviewing. "Spending time reviewing the articles of your colleagues doesn't pay on the individual level as it's voluntary work", explains the representative of French scientific publishers. To attract new profiles and thus increase the capacity to review articles that have been submitted for review, Emmanuelle Jannès-Ober is in favour of peer reviews to be made public and non-anonymous so "reviewers are better valued for their work and this is taken into account in their career assessments".
This position is not broad enough in Lionel Maurel's view because 90% of scientific pre-print manuscripts will eventually be published in a journal. "Peer review has lost its primary function. Its purpose isn't to evaluate quality any more but to rank prestige", laments the deputy director scientific publishing before going on to clarify his point. "Nowadays, scientific publishers no longer sell a certain quality of research – instead they sell prestige which is a commodity of infinite value that's actually unrelated to the service provided but publication costs are based on this. It's no longer a question of deciding whether to publish a paper but rather of choosing where to publish it based on a journal's reputation". Unlike publishers, Lionel Maurel and other contributors believe that the overproduction crisis can only be resolved by reforming the assessment of research to move away from basing this on bibliometric indicators.
- 1Member of the board of directors of the Alliance of French Public Scientific Publishers.
Antoine Petit, the CNRS Chairman and CEO, reminds us that this has been the case at the CNRS since 2021. "The individual assessment of researchers has been completely overhauled in consultation with the CPCN1 to put qualitative arguments back at the core of the assessment process". To help drive the progress of open science practices, the CNRS has put forward four principles to the National Committee (CoNRS) with a dual objective – to promote enhanced recognition of the diversity of the research profession and to base the annual assessment of its scientists on qualitative rather than quantitative criteria. From now on, it is compulsory for assessment reports to include a narrative section describing the importance of a particular discovery rather than being exclusively limited to bibliometric indicators like the impact factor or the H-index. Alain Schuhl mocks this focus on factors like this. "I've never seen anyone be awarded a Nobel Prize because they'd published 2583 articles. They're honoured for the impact of their scientific discoveries".
With this shift away from such indicators, research assessment can now highlight other areas of scientific activity rather than focusing on publications alone. These include communicating on research, outreach activities, cross-disciplinary missions and even investment in the translation of scientific articles, with the latter activity featuring in a series of presentations at the Open Science Day. One was made by Susanna Fiorini, the scientific coordinator of the OPERAS research infrastructure2 dedicated to open science in the humanities and social sciences. One of its services is Mondaecus which works to "bring together translators, proofreaders and experts in different fields to provide high-quality translations". In our era of artificial intelligence, Ms Fiorini calls for "mobilising collective intelligence and disciplinary language expertise around translation". Translation and any other activity that involves the dissemination of scientific knowledge dovetail with the wishes of the CNRS's Deputy CEO for Science who considers that "we'll have to invent new ways of disseminating scientific knowledge".