Comptes Rendus

Use of artificial intelligence

 

Artificial intelligence is transforming the way we conduct research. Today, it offers numerous opportunities, both for the advancement of scientific knowledge and for the daily work of researchers. But it also has many limitations, which mean that it can lead to serious scientific problems and cause abuses that are harmful to society as a whole.

It is up to everyone involved in the publication of Comptes Rendus to ensure that artificial intelligence, particularly generative AI, is an asset rather than a threat.

 

For authors

When used responsibly, artificial intelligence tools can save considerable time and increase efficiency, which is why the Académie des sciences does not generally prohibit their use.

However, it requires users of AI tools to demonstrate the utmost transparency on this subject, in a dedicated section of their article entitled “Uses of generative artificial intelligence.” All uses of AI must also be described in detail in the Cover Letter accompanying the manuscript submission.

 

General instructions

Authorship: Generative AI rools (such as ChatGPT, Copilot or Gemini), must not be listed as co-authors of the publication. These tools cannot be held responsible for the quality, originality, robustness, or integrity of the scientific work presented in the submitted manuscript.

AI as a source: Responses generated by generative AI, and more generally, all products generated by AI tools (e.g., graphics, images, etc.) should not be considered reliable sources of information and should not be referenced in the bibliography of articles.
In the event that an article specifically focuses on the analysis of AI responses, presented and studied as such, and since such responses are by nature not reproducible, the authors must:

  • Mention the tool used, providing all relevant technical information (developers, version used, web browser used, etc.);
  • Provide the prompt used, if necessary in the form of supplementary material associated with the article;
  • Indicate the precise date and time of each response provided.

Generation of figures and visualization techniques: authors are strictly prohibited from using AI tools to generate figures, tables, and other media that enrich the article, due to the high risk of aberrations.

Author responsibility: Any use of artificial intelligence in connection with the submission, whether during the preparation of the research described, during its execution, or during the writing of the submission, must have been supervised and controlled by the authors.
In particular, authors are expected to review, verify, and correct all data and/or codes thus generated, collected, cleaned, analyzed, and/or interpreted, as well as all passages of text thus written. Particular attention should be paid to the bibliographic references cited.

 

Writing the declaration of use

Authors who have used generative artificial intelligence techniques at any stage of their research are invited to use the GAIDeT Declaration Generator (Generative Artificial Intelligence Delegation Taxonomy Declaration Generator) tool to formulate the declaration to be included in a dedicated section of their article entitled “Use of generative artificial intelligence.”
These uses must also be described in detail in the Letter to the Editor accompanying the submission.

 

Failure to comply with transparency requirements

Any suspicion of undeclared use of AI must be immediately reported to the journal editors. Reviewers responsible for evaluating submissions, in particular, are called upon to exercise the utmost vigilance in this regard.

The future of each article can only be decided by the editors-in-chief of the journal concerned (or by the publication directors), after thorough review. In the event of confirmation or strong suspicion of undeclared use of AI:

  • Submissions currently undergoing editorial processing will be rejected;
  • Published articles will be retracted, or an “Expression of Concern” will be published by the editors of the journal.

 

For reviewers

Use of artificial intelligence in the production of evaluation reports

Manuscripts submitted to Comptes Rendus are confidential documents, subject to copyright, and may contain personal and/or sensitive data. Experts are therefore strictly prohibited from uploading manuscripts submitted for their review to artificial intelligence tools for evaluation.

Experts are also prohibited from uploading their own evaluation reports to such AI tools, even for the purpose of improving the style or language level, as documents of this type may also include confidential information and personal or sensitive data (e.g., value judgments).

 

Questions regarding the possible undeclared use of AI

When used responsibly, artificial intelligence tools can save considerable time and increase efficiency, which is why the Académie des sciences does not systematically prohibit authors from using them. However, it requires users of AI tools to be as transparent as possible on this subject, in a dedicated section of their article entitled “Use of artificial intelligence techniques and/or generative artificial intelligence.”

Any reviewer who suspects the use of AI that is not mentioned in a dedicated section of the manuscript they are reviewing is asked to report this to the journal editor as soon as possible. Please note that reviewers are prohibited from using automated AI detection tools on their own.

 

Use of AI by journal editors

The editorial teams at Comptes Rendus are aware of the issues surrounding the use of artificial intelligence in scientific publishing and the bad practices that can be associated with it.
They are also aware of the need to comply with current legislation on copyright and the protection of personal and sensitive data, and therefore refuse to use AI detection tools that are not publicly recognized as being completely secure.

To date, the Académie's editorial teams are likely to use the following AI tools:

  • iThenticate: operated by TurnItIn, this service is integrated into the editorial management software of Comptes Rendus and checks for plagiarism in manuscripts submitted to the Académie des sciences' journals by generating a similarity report that compares the submission to more than 190 million journal articles and content sources from leading scientific publishers, in partnership with Crossref;
  • BibCheck: operated by Inist-CNRS, this web service checks the bibliographic references of an article by verifying their presence in Crossref and ensuring that the articles cited have not been retracted;
  • Antidote: operated by Druide, this writing assistance tool is used exclusively on articles accepted for publication, during an orthographic and typographical correction phase. All corrections suggested by the tool are reviewed and sorted by the journals' editorial teams and proposed to the authors;
  • DeepL: this generative AI-based translation tool is used exclusively on articles accepted for publication to provide a translation of the article's metadata (title, abstract, and keywords) into French or English. These translations are systematically proofread and validated by the authors or editors responsible for the article. Providing multilingual article metadata is part of open science best practices.

The editorial teams of Comptes Rendus are committed to never basing their editorial decisions on AI tool reports.