Community-governed AI for Kauaʻi — for learning, livelihoods, stewardship, and resilience. Owned by the community, grounded in knowledge its holders approve, and guided by the people who call this island home.
A waihona is a repository, a storehouse — in modern Hawaiian usage, even a computer file. The name describes the heart of this project: a protected storehouse of the community’s knowledge, kept on Kauaʻi and held under its holders’ authority.
AI is already transforming how people learn, work, and solve problems. But what if Kauaʻi could offer an alternative to generic, corporate-controlled AI? One grounded in local knowledge, Hawaiian values, and the needs of its people — a locally owned and controlled system people could trust to not sell their data or manipulate their children?
Imagine a keiki (child) struggling with mathematics. Instead of repeating a textbook explanation, an AI tutor connects the lesson to farming, asks questions, adapts to the learner’s understanding, and guides them towards hands-on experience with a local mentor — and prepares them for it.
An ʻulu (breadfruit) farmer uses the same system to diagnose a struggling tree and talk through a repair when equipment breaks down — and to check what buyers are paying before selling, manage orders, plan planting and harvests for CSA (community-supported agriculture) boxes, and keep up with records and accounts. The hours saved go back to the farm and the family. And the farm’s prices, orders, and records stay on a local system the community owns and trusts, never handed to a corporation.
We envision AI that makes learning personal, knowledge interactive, and problem-solving accessible, while strengthening human relationships, perpetuating living traditions, and creating economic opportunity.
Guided by these values, this initiative puts technology in service of people, place, and purpose.
A community-governed AI learning and knowledge platform, powered by community-owned computing infrastructure and open-weight/open-source AI models — models whose underlying files are public, so the community can run and adapt them on its own computers.
It is sovereign because the community owns the computers, holds the data, and approves what the system learns from.
Its learning model is adaptive mastery learning: practice adapts to each learner, and learners advance only by demonstrating knowledge and skills — hands-on, in person, to a teacher or mentor. The platform connects three essential elements:
Community-owned computers on Kauaʻi running open models locally — private by design, and working offline when the island’s connectivity drops.
Adaptive tutors that work alongside teachers and mentors — never instead of them — with mastery verified by people, in person.
Modules authored by the kūpuna, farmers, and practitioners who hold the knowledge — practiced and perpetuated, under their authority.
Community-owned high-performance computers on Kauaʻi will run open-weight/open-source AI models locally, protecting privacy and reducing dependence on outside cloud providers. On those machines, a LoRA adapter — a small trainable layer that adapts the open model to Kauaʻi’s language, values, and teaching style — is trained on conversational examples the Cultural Governance Board approves. Retrieval-augmented generation (RAG) then has the system look up holder-approved knowledge in a protected living archive — the waihona — before it answers, so its answers come from local sources.
That promise covers everything families and children share with the platform. For adults, the same system doubles as a general-purpose AI assistant for everyday work — writing, planning, research, and problem-solving — so the community reaches for its own tool instead of a corporate chatbot.
Solar power and battery backup will run the computers, so daily operation adds no carbon emissions and the system keeps working through outages. Learning materials and AI assistance stay available offline: at participating schools and community sites, learners connect to the platform’s own local network rather than the public internet, so practice continues even when outside connectivity drops.
Interactive AI tutors adapt explanations, examples, and exercises to each learner’s abilities, interests, and local context. They encourage questions, guide problem-solving, and help connect classroom concepts to real-world challenges. Learners use the tutor alongside teachers and mentors, never instead of them: it offers questions and hints before answers, so the thinking — and the productive struggle — stays with the learner.
Teachers get support too: the tutor carries the adaptive practice, holds each learner’s engagement at their level, and shows the teacher exactly where learners are stuck, so teachers spend their time teaching and mentoring. Beyond the screen, AI facilitates mentorship, group projects, and hands-on experiences, bringing learners together with teachers, peers, and community experts. Human mentors guide progress and verify mastery. Learners progress from novice toward mastery at their own pace: each step up is earned by showing the skill in real work — in the field, in the workshop, at community gatherings — before the mentors who taught it.
That structure — sequenced practice that adapts to the learner, immediate feedback, and advancement by demonstrated mastery — is the one behind the strongest credible results in education research. Human tutoring built on it averages 0.37 standard deviations across randomized trials, measured on independent tests — enough to move a learner in the middle of a group to roughly the 64th percentile (Nickow, Oreopoulos & Quan, 2020). Intelligent tutoring systems automate the same structure, and their evaluations carry a warning this page will not hide: across 50 controlled studies, the systems averaged 0.73 standard deviations on tests of the material they taught, but 0.13 on independent standardized tests (Kulik & Fletcher, 2016). The second figure is the honest one — it uses the same yardstick as every other number here. For scale, the median gain across 747 randomized trials of teaching interventions is 0.10 standard deviations, and 0.20 or more counts as large (Kraft, 2020); educational technology programs overall average about 0.15 (Cheung & Slavin, 2013).
What has not worked is the free-standing AI chatbot: in a two-year randomized trial, Khan Academy’s structured practice improved learning while Khanmigo, Khan Academy’s AI tutor, added no measurable gain. So here the AI is a component inside expert-authored modules, with mastery checks and mentor verification — not a stand-alone chatbot, a companion, or a shortcut to answers. A second trial, with more than 6,000 students, found the clearest gains when AI support sat inside a mastery-based workflow rather than beside it (Oreopoulos et al., 2026). Waihona will hold itself to the same standard as the studies above: its modules will be evaluated on independent assessments, with results reported either way.
Sources: Kulik & Fletcher (2016), Review of Educational Research — meta-analysis of 50 controlled studies of intelligent tutoring systems. Kraft (2020), Educational Researcher — effect-size benchmarks from 747 randomized trials. Cheung & Slavin (2013), Educational Research Review — meta-analysis of educational technology applications. Nickow, Oreopoulos & Quan (2020), NBER Working Paper 27476 — meta-analysis of tutoring programs. Oreopoulos & Low (2026), NBER Working Paper 35620 — two-year randomized trial of Khan Academy practice and its Khanmigo AI tutor; and Oreopoulos, Liut, Sungu & Low (2026), NBER Working Paper 35621 — randomized trial of AI support inside a mastery-based practice platform. The 2026 papers are working papers that have not yet completed peer review.
Educators, kūpuna (elders), farmers, and cultural practitioners use AI to transform their expertise into interactive learning modules, from ʻulu cultivation and ahupuaʻa stewardship (caring for a traditional land division from the mountains to the sea) to environmental science, entrepreneurship, and Hawaiian cultural practices. Here is how one module comes to be:
A farmer and a teacher design an ʻulu cultivation module together.
The knowledge the farmer approves enters a protected living archive. Nothing enters without the holder’s approval.
The AI builds questions, examples, and practice from that approved material only, adapting to each learner and showing the teacher where learners get stuck.
The learning finishes in the field, where the farmer watches the learner demonstrate the skill and confirms mastery.
Learners can ask questions, explore scenarios, and work through practical challenges using materials approved by the knowledge holders — the kūpuna, farmers, and practitioners whose knowledge those materials carry. Rather than simply archiving traditional knowledge, the platform helps communities practice and perpetuate living traditions across generations, while cultural authorities retain control over what is shared or protected.
A Kauaʻi Cultural Governance Board, recruited by Kumu Kēhaulani Kekua — a master teacher of hula and cultural leader — will guide community governance, including approving the conversational examples used to train the system, while cultural knowledge holders retain authority over their contributions. Privacy, transparency, safety, and local control are foundational.
Safeguarding children is part of that foundation: children learn from approved material authored by educators, kūpuna, and practitioners; teachers and mentors oversee their progress; and families can see, correct, or delete what the system holds about their child.
The initiative also invests in good local jobs and a stronger island economy, compensating knowledge holders, training local AI technicians and developers, and helping farmers, entrepreneurs, and community organizations innovate and thrive. By keeping technology, expertise, and economic opportunity on Kauaʻi, the project builds lasting community capacity and self-reliance.
over three years — paying local people first: module authors, teachers and mentors, coordinators, hosts, and local technicians, and only then the computers and power resilience that keep the system on Kauaʻi.
The Inter-Nation Cultural Foundation (INCF), a Kauaʻi-based 501(c)(3) nonprofit founded in 1998, leads the initiative and administers its funding. The project’s first application — to Humanity AI’s “Communities Leading on AI” program — was submitted on October 21, 2026. Partner agreements and program details are being formed with participating organizations now, and implementation remains subject to those agreements and available funding.
AI guided by aloha, governed with kuleana, and developed to help the people, culture, and ʻāina of Kauaʻi flourish for generations to come.