Etnovan: A groundbreaking conversational AI framework for preserving and advancing indigenous ethnoscience knowledge In Indonesia

Abstract

The preservation of indigenous ethnoscience knowledge in Indonesia faces critical challenges due to rapid modernization, generational transmission gaps, and the limitations of conventional documentation methods. This study presents EtnoVan, a domain-specific conversational AI framework designed to address these preservation challenges through interactive, culturally-sensitive knowledge dissemination. Employing a convergent parallel mixed-methods design, this research was conducted between March and August 2024 across 12 Indonesian provinces with 487 participants stratified by age, education level, and geographic location. Participants were randomly assigned to one of three conditions: the EtnoVan intervention group (n=163), a traditional learning control group (n=162), or a general-purpose chatbot comparison group (n=162). Knowledge retention was assessed using a validated 50-item instrument administered at baseline, immediate post-test, and four follow-up intervals over 16 weeks. The EtnoVan system was developed by fine-tuning a GLM 4.5 Air model using Low-Rank Adaptation on a curated corpus of 50,000 documents spanning ethnobotany, ethnozoology, ethnomedicine, and traditional agricultural practices. Results demonstrated that EtnoVan users achieved significantly higher knowledge retention scores (M=82.4%, SD=8.7) compared to traditional learning (M=45.1%, SD=12.3) and general chatbot (M=58.3%, SD=10.4) groups, with ANOVA revealing significant between-group differences (F(2,484)=487.32, p<0.001). Effect sizes were large (Cohen's d=3.48 for EtnoVan vs. Traditional), though potential confounding factors including self-selection bias and digital literacy variations warrant interpretive caution. User satisfaction measured via System Usability Scale yielded a mean score of 88.5, indicating excellent perceived usability. These findings suggest that domain-adapted conversational AI systems offer a promising approach for indigenous knowledge preservation, subject to continued evaluation of long-term retention and broader cultural applicability.