IGeLU 2025

Conference

The model of access to documentary resources on the Web has driven the development of information retrieval technologies based on the accumulation of data in centralized computing structures, enabling the construction of appropriate indexes that naturally and efficiently guide search processes. It was precisely this massive accumulation of data, combined with extraordinary computing resources, that allowed Machine Learning methodologies to take off, eventually leading to the development of Large Language Models.

In this presentation, I propose a fundamental reflection on the need for such vast data collections in order to enable relevant cognitive processes. Specifically, I argue that, unlike web search processes, the development of intelligent agents does not require such an accumulation of data. This opens the door to a complete rethinking of Artificial Intelligence methodologies, suggesting that the decentralization of agents designed with specific cognitive capabilities could lead to a new vision of documentary resources, and in particular, of Libraries. Such resources could incorporate privacy by design and form the foundation for new library services supported by intelligent agents operating within a peer-to-peer network.

Finally, I will present the UNaIVERSE system, developed within the University of Siena, for managing the peer-to-peer network that supports communication among intelligent agents.

General / Other