Organized by Soluris, the Municipality of Saint-Pierre-d’Oléron, and the Community of Communes of the Île d'Oléron, the event aimed to decipher AI mechanisms, identify risks, and propose concrete actions for local authorities.
The first part, led by Christelle Plessis and Tristan Fernandez from the Rochefort Digital Collective, explained the workings of everyday AI tools like voice assistants and chatbots. They detailed AI's origins, evolution, and how it generates statistically probable responses. The concept of digital sovereignty was highlighted, particularly for maintaining data control against American, Chinese, or French models.
Practical demonstrations illustrated the importance of formulating precise prompts for relevant results, using tools like ComparIA, recommended by Tristan Fernandez.
The second part, featuring Thomas Llopez from Soluris, addressed AI risks: hallucinations, algorithmic bias, deepfakes, and data leaks. He emphasized that 'AI proposes, humans dispose,' stressing the indispensability of human supervision. Concrete cases, such as prompt injection manipulation in CVs or public tenders, were presented. The RGPD regulatory framework and the AI Act were explained, with a risk-based approach and the need for data anonymization. Establishing an AI charter was proposed to govern usage.
Finally, Matthieu Brient from Les Interconnectés presented practical solutions for integrating AI into local authorities, such as drafting reports or deliberations. He stressed data quality, the need for a roadmap with an AI referent and a charter, and skill sharing. Transparency and exemplary conduct are expected by users, who want to be informed about the use of AI by their communities.
The morning concluded with the understanding that AI is a tool requiring reflection on objectives and existing uses before tool selection, and that governance and agent training are crucial pillars for effective and secure AI implementation.
The Community of Communes of the Île d'Oléron has been experimenting with AI for nearly two years and chose to pilot Delibia from early 2025, aiming to understand its uses and opportunities while considering ethical and environmental challenges.
