Whose Knowledge Counts? Co-Designing Community-Centered AI Auditing Tools with Educators in Hawai`i

Paper Title

Whose Knowledge Counts? Co-Designing Community-Centered AI Auditing Tools with Educators in Hawai'i

Publication Info

• Topic area: Co-designing AI auditing tools for culturally sensitive educational contexts.
• Keywords: AI auditing, generative AI, cultural representation, Indigenous education, Hawaiian culture, co-design, community-centered design, education technology, decolonial computing, participatory design.

Background and Problem

• Problem / challenge: Generative AI systems often misrepresent cultural content, overrepresent Western narratives, and fail to accommodate low-resource languages and Indigenous contexts. Educators lack tools to audit these outputs effectively.
• Significance: Misrepresentations in AI outputs can distort students' understanding of history and culture, particularly in contexts like Hawai‘i, where integrating Hawaiian language and culture into education is mandated.
• Motivation and related work: Prior research has highlighted generative AI's biases and the need for community-driven auditing. However, existing tools fail to address the specific needs of educators in culturally sensitive settings, leaving a gap in designing tools that align with local epistemologies and values.

Solution

• Proposed approach: A community-centered framework for designing AI auditing tools tailored to the needs of educators in Hawai‘i, emphasizing Hawaiian cultural values and practices.
• Novelty:
  1. Identification of five dimensions of cultural misrepresentation in AI outputs specific to Hawaiian education.
  2. Co-designed auditing tool features, including source genealogy, perspective visualization, and flagging problematic outputs.
  3. A reframing of AI auditing as a community-oriented process rather than an individual task.
  4. Design recommendations for embedding Indigenous knowledge systems and ensuring data sovereignty.
• Procedure and key techniques:
  • Conducted four co-design workshops with 22 public school educators in O‘ahu, Hawai‘i.
  • Used design exercises (rapid prototyping and storyboarding) to ideate auditing tools.
  • Analyzed workshop data using reflexive thematic analysis to identify key themes and design priorities.

Results

• Concrete findings:
  • 68.2% of participants used AI in teaching, with 40% employing it for tasks related to Hawaiian culture.
  • Participants identified five cultural misrepresentation concerns: hallucinated outputs, superficial cultural representation, dominance of Western narratives, lack of diversity, and framing Hawaiian culture as historical.
  • Proposed auditing tool features included tracing source genealogy, visualizing perspectives, and flagging harmful outputs.
• Advantage over baselines: Unlike general-purpose auditing tools, the proposed designs embed Hawaiian cultural values, address local epistemologies, and support educators in culturally sensitive contexts.
• Experiments / evaluation:
  • Workshops involved 22 educators across four sessions, using real-world generative AI outputs as design probes.
  • Outputs were analyzed for cultural misrepresentation, and participants proposed tool designs to address identified harms.
• Limitations and future work:
  • Limited to elementary public school educators in O‘ahu; findings may not generalize to other regions, grade levels, or private schools.
  • Focused on text-based generative AI; future work should explore other modalities like images or audio.
  • Further research needed on governance structures for community-led auditing tools and long-term maintenance.

Summary

This study investigates the challenges of using generative AI in culturally sensitive educational contexts, focusing on Hawai‘i's public schools. Through co-design workshops with 22 educators, the authors identified key cultural misrepresentation concerns and proposed auditing tool features, such as tracing knowledge genealogy and visualizing perspectives. The work reframes AI auditing as a community-oriented process, emphasizing the integration of Indigenous knowledge systems and data sovereignty. While the findings are specific to Hawai‘i, the framework offers broader insights for designing community-centered auditing tools in other marginalized contexts.