Knowledge representation is the branch of artificial intelligence concerned with encoding information about the world in formal structures that computational systems can reason over. Rather than simply storing raw data, knowledge representation involves selecting appropriate formalisms, such as ontologies, semantic networks, frames, and logic-based schemas, that capture the relationships, constraints, and categories inherent in a given domain. The field draws on philosophy, linguistics, cognitive science, and information science to bridge the gap between how humans understand concepts and how machines can process them.
Anthropological perspectives enrich the study of knowledge representation by revealing that categorization is never culturally neutral. The taxonomies built into AI systems reflect the worldviews, priorities, and social hierarchies of their designers. Cultural anthropology has long demonstrated that classification systems vary dramatically across societies, from kinship terminologies to ethnobotanical categories. When engineers formalize knowledge into ontologies without accounting for this variation, the resulting systems risk encoding one culture’s assumptions as universal truths, a problem that grows more consequential as AI systems operate across global contexts.
The practical architecture of knowledge representation spans several paradigms. Symbolic AI relies on explicit, human-readable structures like description logics and rule-based systems, where relationships between entities are stated declaratively. Knowledge graphs extend this tradition by modeling entities and their connections as nodes and edges, enabling flexible traversal and inference. More recent hybrid approaches combine symbolic structures with deep learning embeddings, attempting to merge the interpretability of formal representation with the pattern recognition strengths of neural networks.
Matt Artz’s work on ethnographic knowledge semantic data modeling addresses knowledge representation at the intersection of anthropology and AI. His UNESCO paper on moving from machine learning to machine knowing argues that ethnographic observation produces richly contextual knowledge that resists reduction to flat feature vectors. By developing semantic data models grounded in fieldwork findings, the approach preserves the relational and situational texture of cultural knowledge within computational frameworks, treating representation as an interpretive act rather than a purely technical one.
The future of knowledge representation will increasingly confront questions about whose knowledge gets represented and through what conceptual lens. As large language models absorb vast corpora and generate plausible but opaque internal representations, the field faces a tension between statistical models that learn latent structures and engineered ontologies that make their commitments explicit. For anthropologists, this tension is familiar: it mirrors longstanding debates about emic versus etic description, and about whether any representational system can fully capture the meaning-laden, context-dependent nature of human understanding.