Intelligenza artificiale affidabile e sovrana nelle Scienze della Terra

This presentation was delivered at the conference “Geoetica e responsabilità nelle Scienze della Terra”, held at the Istituto Nazionale di Geofisica e Vulcanologia (INGV), Rome, on 12 October 2026. The talk, entitled “Intelligenza artificiale affidabile e sovrana nelle Scienze della Terra”, discusses two complementary dimensions of trustworthy scientific AI. The first part presents the RockGPT approach and the concept of Researcher-Curated Knowledge, showing how Retrieval-Augmented Generation (RAG) and researcher-selected scientific corpora can improve the reliability, provenance, and verifiability of AI-generated answers in the geosciences. The second part broadens the discussion to AI sovereignty, focusing on control over models, data, infrastructure, deployment, and technical skills. Particular attention is given to the European context, the distinction between model sovereignty and deployment sovereignty, open-source AI, the Open Source AI Definition (OSAID), and the need for European-controlled infrastructures and adaptable models. The presentation argues that trustworthy scientific AI requires not only reliable knowledge governance, but also technological agency over the infrastructures used to process and interpret scientific knowledge. Se vuoi, ti preparo anche Title, Abstract, Keywords e Subjects/Communities già ottimizzati per Zenodo.