Epistemological Building Blocks for AI Cognition

This foundational paper proposes epistemological building blocks as fundamental units of knowledge for AI cognition. We define these blocks as the minimal perceptual and conceptual primitives (sensory schemas, categorical concepts, inferential relations, etc.) from which AI can construct richer knowledge. Drawing on philosophy of science and cognitive theory, we argue that an AI’s ability to understand and learn depends on grounding its symbols in sensory experience and organizing them into coherent structures. We situate this proposal against existing knowledge representation (symbolic vs. subsymbolic) and highlight how building-blocks frameworks (from Harnad’s symbol-grounding to hierarchical memory models) can unify these views. Our approach extends current AI by emphasizing intrinsic meaning: symbols must link to sensory prototypes and concepts (the “grounded” building blocks). We outline a method for constructing and integrating these blocks in an architecture, and suggest experiments (e.g. concept-learning tasks with minimal sensory inputs) to validate the framework. We expect that explicitly modeling these primitives will improve AI generalization and explainability. Finally, we discuss implications for cognitive architectures and point to how this builds on related work in embodied cognition and developmental learning.