During the Documation 2025 exhibition in Paris, four experts from the worlds of information, digital archiving and artificial intelligence discussed a simple but strategic question: is AI an ally, a threat or an uncontrollable factor for our long-term preservation systems? Under the leadership of Clémence Jost, editor-in-chief of Archimag magazine, the discussion brought together technical, legal, archival and normative perspectives.
The speakers:
- Morgan Attias, product innovation manager in the field of digital preservation.
- Fabrice Le Gascoin, independent consultant and teacher, specialist in the business uses of artificial intelligence.
- Édouard Vasseur, professor at the École nationale des chartes, expert in archiving and preservation system design.
- Jean-Pierre La Hausse de Lalouvière, president of a professional association committed to the evolution of standards in digital transformation.
Between hope and concern
By way of introduction, Clémence Jost recalls the results of a survey conducted in 2024: 86% of information professionals say they feel fear about AI, but 80% also declare themselves optimistic. This tension clearly reflects the expectations of the sector, caught between the promise of automation and the fear of a decline in the control and reliability of archives.
An asset or an illusion of efficiency?
Morgan Attias opens the debate by emphasising the concrete benefits of AI for digital preservation: fast processing of large volumes, decision support for non-expert users, and automation of tasks such as indexing or pre-sorting. He also mentions the occasional contribution of generative AI in creating scripts or intelligent assistants, provided that their use is supervised.
Fabrice Le Gascoin qualifies this statement. In his view, AIs can produce systemic errors, particularly in the generation of metadata or the classification of documents, resulting in distorted archives. He also warns about the growing complexity of systems, their dependence on technology publishers and the risks this poses to the sustainability of the tools.
Jean-Pierre La Hausse sheds a more inclusive light on the subject, emphasising the role of AI in the democratisation of information. In some countries, such as in Africa, machine translation makes it possible to make public documents accessible in non-digitised local languages, a powerful lever for citizen archiving.
A threat to the reliability of archives?
Édouard Vasseur warns of a growing risk: the confusion between authentic content and generated content. Deepfakes, AI hallucinations or invented texts can be mistaken for reliable sources. Integrated into preservation systems, these documents become references… yet falsified. He insists on the need to keep a human eye on AI productions.
Jean-Pierre La Hausse de Lalouvière reinforces this idea with a concrete case: a lawyer using a consumer cloud had his account deleted after an automatic alert, resulting in the loss of all his archives. This kind of incident underlines the crucial importance of technological sovereignty in the choice of archiving solutions.
Finally, Fabrice Le Gascoin reminds us that AI-generated content detection tools are imperfect, even biased. Some manually produced texts are identified as artificial, while genuine AI texts go unnoticed. This reinforces the idea that only the human eye remains reliable.
How can we prevent AI from becoming the ‘bad guy’?
Morgan Attias emphasises the need for greater traceability: each document must retain a clear chain of provenance, including author, date, tool used and any transformations. He recommends strictly restricting the use of AI in preservation tools, particularly enrichment or translation without human validation.
Jean-Pierre La Hausse de Lalouvière calls for AI to be repositioned as a tool, nothing more. He distinguishes three user profiles: those who reject AI and expose themselves to obsolescence, those who adopt it indiscriminately, and those who integrate it with lucidity, as one lever among others. He cites the example of a hospital that saved time thanks to an AI for pre-analysing X-rays, while maintaining systematic human control.
What future for AI in preservation?
For Morgan Attias, AI is comparable to smartphones in their early days: it seems to be an accessory today, but will quickly become indispensable. He calls for rigorous supervision, both technically and legally, and warns of the environmental impact of generative AI, which should be taken into account in architectural choices.
Fabrice Le Gascoin emphasises that generative AIs, non-deterministic by nature, pose a real problem of reproducibility, a key criterion in archiving. He notes, however, that the gap between the speed of evolution of AIs and that of companies can be an opportunity to experiment with complete clarity.
Édouard Vasseur identifies a major obstacle to the development of AI useful for archiving: the inaccessibility of training corpora for legal reasons. He calls for a supervised opening of these resources to public research, in order to design tools adapted to the memory professions.
Finally, Jean-Pierre La Hausse de Lalouvière concludes with a more geopolitical vision: mastering AI is a strategic issue for Europe, which must retain its sovereignty over critical technologies. He also refers to the imminent arrival of quantum computing, which will in turn upset the balance, including in conservation.
Conclusion: remain curious, critical and vigilant
The speakers share a common conviction: AI is neither good nor bad in itself. It all depends on how it is conceived, used and controlled. In an environment in which technologies are constantly evolving, preservation professionals must remain open, test, correct, train continuously and, above all, keep a critical eye on the results produced.
Because beyond the tools, it is indeed a certain idea of memory and human responsibility that must be preserved.