Between predation and disembodiment, scientific publishing to the test of AI
- Cet article est une traduction de :
- Entre prédation et désincarnation, l'édition scientifique à l'épreuve de l'IA
- Autre(s) traduction(s) de cet article :
- Entre la depredación y la desnaturalización: la edición científica a prueba de la IA
- 人工智能对科学出版的挑战:捕食与脱离实际情境之间
- Μεταξύ αρπαγής και 'αποενσωμάτωσης', η επιστημονική έκδοση απέναντι στην πρόκληση της τεχνητής νοημοσύνης
Texte intégral
1While the summit for action on artificial intelligence has just come to an end in Paris, it is important to examine what changes AI, and more particularly generative artificial intelligence, is likely to introduce in the practices of knowledge sharing, scientific writing and peer reviewing.
2Extremely easy to use, conversational robots are, for the most, accessible free of charge to all citizens. For researchers, they have many advantages such as editorial assistance, the creation of summaries and abstracts, translation, data processing, etc. However, we cannot misunderstand or forget how they work. The term "intelligence" is in this case misleading. We are dealing with algorithms associated with computing powers that certainly exceed human intelligence in terms of volume of data processed and speed, but their responses result from a calculation of association probabilities and not from a qualitative analysis of sources. Based on a neural network architecture, the generative AI model does not recognize the intrinsic value, true or false, or the semantic meaning of the content. In addition, in the absence of being able to identify the knowledge base with which the model was trained, no information generated can be scientifically validated. It is the duty of the scientist, author or evaluator to verify everything, including the veracity of the bibliographic references generated, which are sometimes apocryphal.
- 1 Van Noorden R., 2023, "More than 10,000 research papers were retracted in 2023 — a new record", Nat (...)
3In 2023, the withdrawal of more than 10,000 articles from journals published by famous scientific publishers is, according to experts, only the "tip of the iceberg"1. These articles had successfully passed the review stage, but it turned out that some of them had been generated with the assistance of AI in defiance of ethical rules and that another part had been produced in "article factories" from predatory journals. These predatory journals have flooded scientific communication by tens of thousands thanks to open access. And despite the retractions of these articles, references and quotes remain present on the web, in other articles, in the results of conversational robots... and even in a priori serious search engines such as Google Scholar.
4For unscrupulous evaluators, it is convenient to copy and paste the texts they have in reading and use robots to write evaluation reports in their place. These reports are totally disembodied, have no scientific validity, but can lead to the acceptance of publication. Another Pandora's box opens here again since it is very likely that the content deposited in the conversational chat is in turn stored in the robot's knowledge base and can thus be restored in new generations of content...
5Exposed like all scientific publishing to these technological shocks, Cybergeo opts for confidence in human intelligence and in the ethics of the scientist in order to guarantee the rigor and validity of the articles it publishes. All authors and evaluators undertake to respect its ethical charter and its instructions to authors. It does not exclude the use of AI, but asks authors to indicate at the beginning of their text, if applicable, the nature of AI assistance and to certify that they have verified all the results provided as well as their bibliographic references. Not recognizing it can be considered plagiarism and lead to the withdrawal of the article. Evaluators are invited to look closely at the contents, data and bibliographic references and to report any inconsistencies. The use of AI is also firmly prohibited for evaluation reports, pending the release of a reliable detection tool that does not, by default, memorize the content deposited therein.
6In a previous editorial, the CybergeoNetworks2 project was presented as an original use of AI in the service of scientific publishing. There is no doubt that the journal can accommodate in the future many other achievements demonstrating all the creative richness that this new tool can bring to well-led research.
Notes
1 Van Noorden R., 2023, "More than 10,000 research papers were retracted in 2023 — a new record", Nature, Vol.624, N°7992, 479‑481. https://www.nature.com/articles/d41586-023-03974-8
Haut de pagePour citer cet article
Référence électronique
Christine Kosmopoulos, « Between predation and disembodiment, scientific publishing to the test of AI », Cybergeo: European Journal of Geography [En ligne], Éditoriaux, mis en ligne le 13 février 2025, consulté le 27 mars 2025. URL : http://journals.openedition.org/cybergeo/41702
Haut de pageDroits d’auteur
Le texte seul est utilisable sous licence CC BY 4.0. Les autres éléments (illustrations, fichiers annexes importés) sont « Tous droits réservés », sauf mention contraire.
Haut de page