Artificial intelligence
Since its inception, Cairn.info has built a vast digital corpus of highly structured scholarly publications. With the recent proliferation of artificial intelligence technologies comes a dual responsibility: mastering such technology to maximize the discoverability of these works, and protecting researchers, publishers, and readers from any attempts by major tech companies to monopolize this wealth of knowledge.
A sovereign infrastructure
In a context of constantly evolving AI technologies and uses, Cairn.info has chosen to equip itself with an autonomous and modular hardware and software architecture, hosted entirely locally. The goal: to be able to work on the latest LLMs, while protecting the publications entrusted to it from any risk of external predation.
This architecture consists of a dedicated network of servers and GPU cards, as well as a suite of software tools for testing and deploying a wide variety of open source language models, both large and small.
Partners
Kairntech, Isako and Pythagoria
SophIA, a RAG to meet the needs of students and researchers
Supported by the Centre National du Livre, SophIA is an alternative to traditional search engines, allowing users to formulate questions in natural language. It differs from other existing tools in two ways: firstly, it is based solely on validated scientific content published on Cairn.info; secondly, it offers direct access to excerpts from these publications, highlighting researchers and their various responses to the questions asked. Rather than attempting a risky synthesis and offering approximate paraphrases, users can discover the scientific exchanges surrounding the issue in question in context (chronological, geographical, disciplinary, etc.).
-
RAG (Retrieval-Augmented Generation) combines two functions: searching for information in a reliable corpus and generating text. AI begins by querying a database or content library, then formulates a clear, contextualized response. This approach guarantees sourced, accurate, and more reliable responses than those produced by a “pure” model.
SophIA will be available at the end of 2025 to Cairn Pro subscribers (psychology, social work, education sciences), and its corpus will then be extended to all disciplines published on the platform.
AI promoting accessibility
Like its publishing partners, Cairn.info is subject to the Digital Accessibility Act, which stems from the European Accessibility Act (EAA).
The new Cairn.info portal has been audited in this regard, and corrections will be made at the start of the 2025 academic year to ensure that the site complies with the recommendations of the RGAA (Référentiel Général d’Amélioration de l’Accessibilité, or General Accessibility Improvement Framework). With regard to the accessibility of publications, alternative texts to images are already being generated for all new publications posted online. A pipeline automatically distinguishes between image types and selects the optimal language model or algorithmic tool for each one to generate an alternative text version.
-
General Accessibility Improvement Framework (RGAA)
French regulatory framework defining the technical criteria for making a website accessible to people with disabilities. It is based on the WCAG, adapting them to the French context. Legal requirement for public websites in France.
https://accessibilite.numerique.gouv.fr
Ethical guidelines
These projects are carried out in compliance with the guidelines that Cairn.info has established for its use of artificial intelligence, pledging to:
- refrain from transmitting partner content to LLMs operated by major AI players;
- protect partner content from a technological (Datadome) and legal (TDMRep) standpoint;
- use generative AI to facilitate access to the academic corpus and not as an alternative to reading the original texts;
- involve partners (authors, publishers, librarians) in the implementation of AI projects in a transparent and collaborative manner; and
- prioritize digital sobriety by selecting the smallest language model capable of solving each problem.
