Collaborative Neurosymbolic AI System Engineering through the BEAM Framework

Neurosymbolic AI, which combines the strengths of neural networks and symbolic reasoning, holds great promise for building interpretable and context-aware AI systems. However, the complexity of these systems introduces significant challenges, particularly in understanding their architecture and accommodating diverse stakeholder perspectives. To address this, the Boxology Extended Annotation Model (BEAM) has been developed as a visual and formal framework for representing AI systems. BEAM extends the original Boxology notation to support multi-perspective representations, enhancing documentation and analysis of AI systems throughout the AI lifecycle. This project aims to further develop BEAM through creation of a shared international knowledge base and exploring possible extensions of the BEAM framework, such as automatic extraction of BEAM models from scientific and software artifacts.
Furthermore, we plan to conduct cross-institutional evaluation to assess the framework’s usability and effectiveness.