From Algorithmic Voice to Community Voice: Authentic Narratives and a Community-Based LLM for Learning

• In Event: Teaching and Learning in Sociology Section Roundtables
  In Refereed Roundtable (90 minute): Roundtable 1: AI and Technology

Abstract

Large language models (LLMs) such as ChatGPT and Gemini have rapidly entered educational settings, offering students fluent but decontextualized approaches to learning. While these systems excel at delivering information at an abstract level, they overlook the social and interpretive dimensions of learning emphasized by constructivist traditions. This paper proposes a Community-Based Large Language Model (CB-LLM) that embeds the voices and narratives of a real learning community into an AI system. Drawing on constructivist, social, and situated learning theories, we argue that learning develops through interpretive engagement with others’ experiences and the contexts in which knowledge is lived. To demonstrate this approach, we built a Community-Based LLM (CB-LLM) incorporating reflective essays written by sociology students who applied theoretical concepts to their own experiences. Using a corpus of 500 anonymized student assignments, the CB-LLM produces outputs that integrate three elements: students’ narratives, the application of sociological concepts, and clear conceptual definitions. In our qualitative comparison with Gemini, the CB-LLM generated more socially situated explanations and, even when Gemini was forced into the same narrative format, the CB-LLM remained distinctive for the authenticity and “bumpiness” of real student voice that prompted deeper interpretation. The paper discusses three key implications: (1) pedagogical, showing how peer narratives foster deeper engagement and belonging; (2) institutional, illustrating how universities can build bespoke, community-specific models that enhance educational autonomy; and (3) political-economic, framing the CB-LLM as a form of data dignity that decentralizes AI and allows communities to own and benefit from their collective knowledge. Together, these findings reimagine AI not as a detached information tool but as a collaborative infrastructure for meaning making in education.

Authors

• Sang Teck Oh, University of North Carolina-Wilmington

• Christopher Shane Elliott, UNCW

• Douglas J. Engelman, University of North Carolina Wilmington

• Luke Christopher Butler