An Integrative Conceptual Review of Federated Learning Ontology: Reconceptualizing the Global Model as Distributed Intelligence

Federated Learning (FL) is widely recognized as a privacy-preserving architecture, yet the ontological status of the "global model" within this distributed system remains under-theorized. This study challenges the purely technical view by reinterpreting FL as an ontological regime of distributed intelligence, where the model exists as a causally effective informational entity. By synthesizing Informational Realism and Actor-Network Theory into a unified Ontological Mediation Framework (OMF), we argue that the global model acquires reality through continuous mediation among algorithms, local data, and institutional actors. To contextualize this framework, the study examines the Federated Tumor Segmentation (FeTS) Initiative, illustrating how a non-local model exerts tangible causal influence within a real-world medical consortium. Furthermore, the paper proposes two normative instruments, specifically the Ontological Impact Assessment (OIA) and the Ontological Audit Framework (OAF), to translate these philosophical insights into practical governance mechanisms for transparency and accountability. This research contributes to the foundations of Responsible AI by positioning Federated Learning not merely as a computational tool, but as a socio-technical process where existence, knowledge, and ethics are intrinsically linked.