Indigenous AI
Motivation:
The rapid advancement of Large Language Models (LLMs) offers the potential to reshape public understanding of indigenous histories and knowledge domains. Recent research explores the use of AI technologies to interpret and convey cultural heritage, addressing a crucial need for preserving indigenous culture and linguistic diversity.
Problem:
The proficiency of advanced LLMs raises concerns about accurately representing and processing data from traditionally marginalized sources, particularly indigenous languages and cultures. This situation has led to heightened apprehensions over data sovereignty issues, emphasizing the importance of recognizing and respecting indigenous communities' rights in the development and application of AI technologies.
Method:
Co-design Methodology: Our research embraced a co-design approach, engaging directly with indigenous communities to develop an AI chatbot using large language models. This methodology facilitated a balanced partnership between users and designers, integrating stakeholders as equal partners in the design process.
Iteration 1 - Investigating Indigenous Perspectives: Fieldwork with the Omaha Tribe involved think-aloud interviews to explore ChatGPT's utility in tribal emergency management. Insights highlighted the need for culturally specific responses and the adoption of the Retrieval-Augmented Generation (RAG) method to integrate indigenous cultural information into language models.
Iteration 2 - Embedding Indigenous Cultural Data: A PDF compilation from the Sac and Fox tribe was uploaded to AskYourPDF, enabling the chatbot to generate contextually relevant responses. A co-design session further refined the chatbot's features, emphasizing accessibility, cultural sensitivity, and privacy.
Iteration 3 - Developing TribalLLM: TribalLLM was developed to allow users to report flood damage, incorporating data from multiple tribes into a local database. An online interview study gathered feedback on TribalLLM's usability and cultural appropriateness, informing further refinement of the tool.
This system screenshot illustrates the co-design system interface during the first iteration. This interface allowed users to pose questions, generating responses based on a provided PDF document containing tribal information. The system also included a feature to cite sources for these responses, enhancing the transparency and credibility of the information presented
This figure displays a screenshot of the TribalLLM chatbot, which integrates data from multiple tribal nations into a large language model. Users can input various prompts to elicit responses from the application, demonstrating the chatbot’s capacity to access and process the embedded tribal data.
Findings:
AI-Generated Responses and the Lens of Native Lands
Diverse emergency scenarios among tribes necessitate localized information for effective response.
Accessibility challenges for elders and individuals with disabilities in using the chatbot during emergencies.
Concerns about resource navigation, including bureaucratic hurdles and lack of awareness about obligations and resources.
Sovereignty and Empowerment through Education
Importance of representing tribal sovereignty accurately in chatbot responses.
Addressing stereotypes and promoting cultural awareness through education and combatting discrimination.
Empowering tribal governments through education on legislative processes and advocacy.
Cultural Preservation and Indigenous Data
Significance of oral traditions and storytelling in preserving tribal culture.
Language preservation efforts to integrate tribal languages into chatbot responses.
Recognizing and respecting cultural differences among tribes in chatbot interactions.
Right to Self-Determination
Importance of reliable data sources and building trust with tribal communities.
Challenges in obtaining authentic tribal knowledge due to historical trauma and lack of written documentation.
Building trust through personal connections and gradual introduction of technology to tribal members.
Broader Impact:
Cultural Sensitivity: Enhances chatbot's cultural authenticity, aligns with indigenous data sovereignty.
Empowerment: Supports indigenous communities with resource guidelines, educational aid, and business support.
Data Sovereignty: Promotes indigenous data control, proposes local database framework.
Cultural Preservation: Advocates for digital preservation respecting indigenous autonomy.
Community Engagement: Stresses long-term collaboration and tailored education for inclusivity.
Respect for Autonomy: Demonstrates commitment to respectful integration of AI in indigenous contexts.