Grounding large language models in an anthropological knowledge graph: A neuro-symbolic approach

As artificial intelligence (AI) systems increasingly mediate access to knowledge, anthropologists face questions about how to engage with these technologies constructively. In this article I demonstrate how pairing large language models (LLMs) with knowledge graphs, an approach called neuro-symbolic AI, can transform how we engage with our own disciplinary scholarship. I constructed the SfAA Bibliographic Graph, representing 84 years of publishing (1941–2025) from the Society for Applied Anthropology’s two journals, and conducted comparative tests presenting identical questions to an LLM alone versus the same LLM integrated with the graph. The results reveal systematic differences. Whereas the LLM alone acknowledges limitations or generates plausible but unverifiable predictions, the graph-grounded system provides precise findings traced through collaboration networks, institutional affiliations, and temporal patterns. The value lies in their interplay: the graph provides deterministic anchors whereas the LLM contributes interpretive context, producing insights neither component could generate alone. For anthropologists, this integration opens new possibilities for how we discover, synthesize, and build upon decades of scholarship, while raising questions about who builds knowledge infrastructure, whose knowledge gets represented, and the invisible labor required to sustain these systems.