Towards a Web Standard for Neuro-Symbolic Integration and Knowledge Representation Using Model Cards
Paola Di Maio · 2023
Neural symbolic integration, in addition to its classical role in bridging symbolic and subsymbolic approaches in artificial intelligence (AI), can be leveraged to connect the shared conceptual structures in data science, the semantic web and AI, as well as to provide explicit and shared knowledge representation (KR) mechanisms to support the explainability, reliability and reproducibility of machine learning (ML). This chapter introduces and explains a novel approach to neuro-symbolic integration using a model card in the context of intelligent systems design, and presents for the first time a pragmatic approach to neuro-symbolism in AI using model cards. A systems development approach is undertaken and consideration is given to early efforts where unified languages and methods were first proposed to tackle disparate integration challenges in AI and where the relationship between knowledge representation and connectionism has been long researched. In addition to bridging symbolic and connectionist AI, integrated neuro-symbolic knowledge representation has a novel role bridging disparate but closely related knowledge domains, such as data science, semantic web and AI, considering advances in neuromorphic engineering. A model card is proposed to achieve integrated neural-symbolic knowledge representation across domains. Application scenarios in neuroscience are used to describe use cases showing how the model cards can be used. The chapter points to the development and publication of a possible web standard for neuro-symbolic integration using model cards.