Community-Verified AI: A Human-in-the-Loop Framework for Preserving Regional Cultural Knowledge
Proposes a verification methodology pairing domain experts (cultural institutions, practitioners) with AI-generated content to reduce hallucination risk and ensure trustworthy representation of niche cultural knowledge.
Abstract
We propose a community-verified AI framework that pairs domain experts — cultural institution curators, art practitioners, and historians — with AI-generated content to reduce hallucination risk and ensure trustworthy representation of regional cultural knowledge. The framework addresses the unique challenge of verifying claims about niche cultural practices where ground truth is often oral, undocumented, or contested.
1. Introduction
LLMs hallucinate. In general domains, this produces incorrect but often detectable errors. In cultural heritage contexts, hallucinations are particularly dangerous: a fabricated claim about a sacred ritual or misattributed art form origin can propagate misinformation about living traditions. Standard fact-checking approaches fail when knowledge exists primarily in oral or community memory.
2. The Cultural Hallucination Problem
2.1 Types of Cultural Hallucination
- Fabricated historical events
- Misattributed art form origins
- Conflated regional traditions
- Invented terminology
2.2 Why Standard Approaches Fail
3. Proposed Framework
3.1 Expert Panel Design
3.2 Verification Workflow
3.3 Confidence Scoring
3.4 Feedback Loops
4. Expert Categories
- Cultural institution curators
- Practicing artists and performers
- Historians and researchers
- Community elders and oral tradition keepers