J . -M . Bertho et al . : Radioprotection 2024 , 59 ( 3 ), 152 – 154 153
médicales du comité international des rédacteurs de publications médicales – International Commission of Medical Journal Editors ( ICMJE ) accessibles via le lien suivant : http :// www . icmje . org / recommendations . Concernant l ’ utilisation des logiciels d ’ IA , il faut distinguer les outils d ’ assistance ( faire des suggestions , des corrections et des améliorations du contenu que vous avez vous-même créé ) et les outils d ’ IA générative ( ChatGPT , Dall-e , ...) qui produisent du contenu . Or , l ’ utilisation de l ’ IA relève entièrement de la responsabilité des auteurs , comme c ’ est le cas pour l ’ utilisation de n ’ importe quel autre outil . Les auteurs sont dorénavant tenus d ’ informer le lecteur de tout contenu généré par une IA générative apparaissant dans leur travail ( y compris le texte , les images ou les traductions ), selon les modalités décrites dans les instructions aux auteurs ( https :// www . radioprotection . org / fr / pour-les-auteurs / instructions-aux-auteurs ). Ces informations obligatoires permettront à l ’ équipe éditoriale de prendre une décision de publication éclairée .
De même que pour le plagiat qui a quasiment disparu , du moins dans les manuscrits reçus par notre journal car le risque de détection est bien trop grand pour les tricheurs , gageons que cette demande de signalement de l ’ utilisation de l ’ IA permettra de réduire les utilisations abusives ou à des fins de malversations de l ’ IA dans la publication scientifique . En attendant , la montée en puissance de l ’ utilisation de ces outils d ’ IA nécessite une vigilance encore plus grande , non seulement de la part des éditeurs , mais aussi des relecteurs de manuscrits , qui utiliseront des logiciels de détection d ’ IA s ’ ils en ressentent le besoin .
For or against the use of artificial intelligence to write scientific articles submitted to Radioprotection
***
We recently received a literature review manuscript so well finished that it seemed to be the result of very careful work . However , various errors led us to think that this article could be the result of using a generative artificial intelligence ( AI ) algorithm . We then used a detection software , which confirmed a very high probability of using generative AI . This submission , the first that we detected within the Radioprotection editorial board , leads us to address the issue of using AI in a scientific publication .
Our first instinctive reaction was to think that this was cheating and that this type of behavior should be banned . Indeed , as members of the editorial committee or reviewers , our expertise , allowing us to evaluate the quality of the manuscripts submitted to Radioprotection , is acquired in part thanks to our publications . Unfortunately , AI today allows novices ( non-experts ) to provide manuscripts without expertise ( which is fraudulent ) and ultimately to be unduly considered experts . In this context , using AI tools , whatever they may be , raises serious questions for us . The Radioprotection editorial committee , in conjunction with our publisher EDP Sciences , therefore looked into this question .
The use of AI tools is in line with scientific and technical developments in our society and will certainly impact radiation protection ( Andresz et al ., 2022 ). The use of AI technologies has been proposed by the NERIS platform to develop advanced tools for emergency preparedness , response and remediation in the event of a nuclear accident ( Bexon et al ., 2023 ). The International Commission on Radiological Protection ( ICRP ) considers the use of AI tools in the current review of the radiation protection system ( Clement et al ., 2021 ). In medical imaging , many solutions based on deep learning methods are marketed for segmentation , classification or identification tasks on x-ray , CT or MRI images , for example to detect fractures or estimate bone age in children , etc ... These new software have real clinical benefits in reducing error rates , improving workflow and reducing radiologist fatigue ( Ruitenbeek et al ., 2024 ; Kelly et al ., 2022 ). Additionally , in the past 5 years , deep learning CT image reconstruction has been introduced to improve image quality and reduce ionizing radiation dose compared with conventional reconstruction techniques ( Zhang et Seeram , 2020 ). These techniques must still be validated in a multicenter manner for greater reliability .
Using AI tools for data production or for writing a manuscript therefore seems inevitable . Each time progress has been made , there have been good and bad uses , but above all good or bad intentions . The example of the use of ionizing radiation is characteristic , almost caricature , from this point of view . It is therefore the user ’ s intention that must be taken into account . Does he simply wish to use a powerful tool to enable him to improve his scientific production , or does he wish to use these tools for scientific or other malicious purposes ?
An interesting parallel can be made with plagiarism , another aspect of scientific integrity . The methods , the way of presenting a corpus of data or even the writing of certain parts of articles cannot avoid a certain similarity , sometimes significant , with existing writings . Who has never copied and pasted a paragraph from materials and methods ? In this case , it is not a question of plagiarism , nor even of an intention to plagiarize , but of a difficulty in expressing the same thing with different words . On the other hand , the fact of voluntarily copying results , interpretations of results and even the conclusions of a scientific article to take ownership of the authorship is clearly plagiarism and must be banned from practice . An essential element results from this example : plagiarism is intentional .
The same goes for using AI tools in scientific publishing . It is not the use of the tool that needs to be questioned . In any case , like any technological progress , its use becomes inevitable , and those who do not use them will be overtaken by the evolution of the increased quality of manuscripts generated with the help of AI . On the other hand , what is important for publishers is to ensure that the use of AI in scientific work is reported , like any other tool used in scientific work , whether it be chemical product , a strain of mouse , or statistical analysis software .