Integrating Perspectives: Standpoint Logic for Knowledge Representation – SPaRK

Artificial intelligence initiatives like the semantic web enable the linking of machine-readable data from various sources and the exploitation of knowledge representation (KR) artefacts, such as ontologies and knowledge graphs, to support advanced automated reasoning tasks. Coupled with the vast amount of data now available, this presents significant opportunities for collaboration, research and innovation across multiple sectors. In healthcare, for instance, integrating patient data from different providers can improve treatment outcomes, personalise care, and enhance coordination.

While the semantic web supports knowledge sharing, a remaining challenge is knowledge interoperability, i.e., seamlessly integrating the knowledge embedded in the diverse data sources available. This is difficult because KR artefacts implicitly reflect the context and perspectives of their creators, often leading to conflicts when different sources are merged. To overcome this, the coordinator of SPaRK introduced Standpoint Logic (SL), a flexible multi-modal logic that allows for the representation of knowledge relative to different, potentially conflicting viewpoints. SL is designed to support knowledge integration and reasoning across varied contexts.

The SPaRK project addresses the knowledge interoperability challenge by proposing SL as a general framework for knowledge integration. To achieve this, we will explore the computational costs of extending KR languages with SL, develop efficient reasoning tools, and establish mechanisms to manage dynamic distributed knowledge networks like the semantic web with the framework. By enabling the integration of diverse knowledge sources while maintaining their expressiveness, SPaRK will enhance the interoperability of AI systems, particularly in critical areas like healthcare or energy. This will allow users to combine and reason with the available data, resulting in more informed decision-making in different domains.