Send message to Contributor

Contribution short abstract

How do we preserve ethnographic data integrity with LLMs? This contribution proposes RAG plus GitHub infrastructure enabling LLMs as faithful assistants while maintaining field notes as ground truth, documenting interpretive circularity, and preserving ethical control over sensitive data.

Contribution long abstract

Computational anthropology (CA) faces a foundational tension: how can we leverage artificial intelligence and LLMs to advance qualitative, ethnographic research without reducing lived experience to statistical patterns or introducing hallucinations that compromise fieldwork integrity? This contribution proposes a methodological infrastructure that preserves ethnographic data fidelity while enabling collaborative, computational augmentation of qualitative analysis.

I argue that field notes must remain the epistemological ground truth in human-centered, AI-assisted research. By embedding researcher-collected ethnographic data within RAG systems, LLMs function as faithful interpretive assistants rather than autonomous knowledge generators; they consistently retrieve and contextualize field observations without fabrication. This approach addresses a challenge in CA: enabling quantitative scaling and pattern detection across qualitative datasets while maintaining the thick description, contextual sensitivity, and ethical accountability.

Complementing RAG, collaborative version control systems (such as GitHub) create transparent audit trails for interpretive circularity. As anthropologists iteratively refine codes, annotations, and analytical frameworks, version control preserves the genealogy of interpretation, enabling reflexivity about how field data transform into ethnographic insight. This dual infrastructure addresses the methodological polarisation the roundtable identifies as central to CA's fragmentation, distributing analytical authority while maintaining fidelity to the original field record.

This synthesis avoids both reductionism and particularism: quantitative capabilities emerge from qualitative foundations, and computational tools serve as transparent, auditable extensions of human interpretive work rather than autonomous knowledge engines. Field notes remain irreducible; algorithms remain traceable. This positions CA not as a methodological divide, but as a practice of disciplined, collective knowledge production grounded in the ethnographers' experiences documented.