ISIDORE is a platform dedicated to research in the humanities and social sciences (HSS), allowing users to explore a vast collection of scientific documents. By using ISIDORE, researchers and students can access articles, theses, reports, books, and many other types of documents. For the past 14 years, the ISIDORE platform has offered advanced search features, such as the ability to filter results by document type, publication date, or discipline, while enabling full-text search across the millions of resources it contains. Additionally, ISIDORE provides data visualization and analysis tools, helping users better understand trends and developments in the field of HSS. By integrating ISIDORE into their research process, users can not only enrich their work with reliable and relevant sources but also contribute to the dissemination and promotion of HSS research.

In the era of widespread artificial intelligence (AI), the main challenge for HSS researchers using these technologies lies in the control and understanding they can—and must—have over them. In this context, the systematic association between large language models and commercial entities, through a classic antonomasia phenomenon in GAFAM services, is particularly concerning: “ChatGPT” has come to represent AI. However, this association raises deep ethical issues: data sharing and legal problems, lack of understanding of underlying technical challenges, loss of individual autonomy, ecological cost, etc.

The ISIDORE 2030 program aims to overhaul the search engine by integrating AI components. Unlike services developed by GAFAM, ISIDORE 2030 intends to put the researcher back at the heart of the AI control system through three key concepts: explainability of AI, interpretability of generations, and adaptability of AI.

  • Explainability: By combining classical text-mining methods (information retrieval, classifications, regressions, clustering, etc.) with AI methods and highlighting parameters, metrics, and corpora.
  • Interpretability: By developing tools to retrieve the sources on which AI relies to generate responses (RAG, in particular) and using frugal, open-source models.
  • Adaptability: By fine-tuning AI models for specific tasks or disciplines. This fine-tuning can also be based on the researcher’s profile, provided it is transparent and adjustable by the researcher.

Thus, in contrast to the invisibilization of the socio-technical processes at work in the AI we encounter daily, our poster, presented at the Humanistica 2025 conference in Dakar (Senegal), aims to show the evolution of our thinking on the development of research assistants adapted and co-created by HSS researchers.

Bibliography (french and english)

  • POUYLLAU, S. (2024, octobre 4). ISIDORE 2030 : adapter les IA aux besoins de la recherche de documents et de données en SHS. Conférence au GF2i (GF2i), PARIS. HN LAB. https://doi.org/10.5281/zenodo.13892964
  • POUYLLAU, S., FACI, A., SILVESTRE DE SACY, A., & MARONET, L. (2024). ISIDORE 2030 : de l’IA de traitement au Retrieval Augmented Generation pour les SHS (1.0). HN Lab. https://doi.org/10.5281/zenodo.14019295
  • POUYLLAU, S. (2022). Refonder ISIDORE (2.0). Zenodo. https://doi.org/10.5281/zenodo.8086278
  • SILVESTRE DE SACY, A., FACI, A., MARONET, L., & POUYLLAU, S. (2024). Note sur l’expérience de l’IA au sein de l’Huma-Num Lab (huma-num version) (1.1). ACFAS 2024 (ACFAS), Ottawa. HN Lab. https://doi.org/10.5281/zenodo.10846773
  • POUYLLAU, S., MINEL, J.-L., CAPELLI, L., SAURET, N., BUNEL, M., BAUDE, O., JOUGUET, H., BUSONERA, P., & DESSEIGNE, A. (2021). ISIDORE celebrates its 10th anniversary. Zenodo. https://doi.org/10.5281/zenodo.5700008
  • Silvestre de Sacy, A., Faci, A., Pouyllau, S., & Maronet, L. (2024, octobre 18). Pre-targeted-RAG - Retrieval Augmented Generation sur des groupes pré-ciblés de communautés d’articles de recherche. ColDoc, Université Paris-Nanterre. Zenodo. https://doi.org/10.5281/zenodo.13950650