From Data Spaces to Agent Spaces. Polleres, A., Rincon-Yanez, D., Anjomshoaa, A., Dobriy, D., & Sallinger, E. In The Fourth International Workshop on Semantics in Dataspaces (SDS 2026), co-located with ESWC, Dubrovnik, Croatia, May, 2026. to appear
From Data Spaces to Agent Spaces [pdf]Paper  abstract   bibtex   
Data Spaces have been incorporated into Europe's Digital Strategy as an alternative, ledger-based approach to sharing data in a decentralised, self-sovereign manner. As technological developments focus on effective, decentralised data sharing, the integration with (tool-enabled) LLMs and emerging Agentic AI remains unclear. In this position paper, we outline how the core ideas of Data Spaces should be thought forward, towards Agent Spaces. We conclude that such an integration will eventually need to rely on hybrid (``bilateral'') AI approaches that combine the base idea of self-sovereign data, with (1) self-sovereign Agentic LLMs (incl.\ tools/services), building on (2) Linked Knowledge Graphs, and adding (3) symbolic guardrails that govern the interaction of agents. Such ``bilateral'' (agentic and neuro-symbolic) AI approaches integrate \emphsymbolic rules, constraints, policies and planning capabilities with \emphsubsymbolic LLMs and learning capabilities.

Downloads: 0