When or If AI Meets Indigenous Knowledge?
An Ideation Series
I have been wrestling with something for a while now, and I keep coming back to the same two thoughts pulling in opposite directions.
The first: Native-serving organizations need help we cannot always afford. Grants, reporting, evaluation, financial management, governance, organizational development, the everyday work of keeping an organization healthy and funded. Technical assistance exists for this, but it is not always available when a small staff, stretched thin, needs it at nine o'clock at night before a deadline. I have seen what that looks like. I have lived it. And I have watched AI do things in a minute that used to take me an afternoon.
The second thought is harder. The more I watch the civic world race into AI- public services, grant administration, data analysis, decision support- the more I wonder whether Native communities will once again be the last ones at the table, asked to live under rules we never helped shape. And I do not know how to hold both of these at once. Maybe that is the point of writing this down.
I do not think we get to opt out of this moment. I just do not want my community to end up governed by tools we never shaped.
So this is the question I keep circling: who decides when — or whether — AI meets Indigenous knowledge?
A Two-Sided Problem
Governments and publicly funded systems are adopting AI for everything from drafting and summarization to data analysis and decision support. That is not coming; it is here. And Native-serving organizations, Tribal nations, Alaska Native communities, Native Hawaiian organizations, Indigenous communities of U.S. territories, hold something these systems keep reaching for: language materials, oral histories, place-based knowledge, community-generated data, cultural information, program records. Much of it is governed by Tribal authority, community protocols, organizational agreements, or restrictions on reuse.
Here is the part I keep turning over. Information can be digitally available without being appropriate for unrestricted AI processing, model training, summarization, or secondary use. Just because it is findable does not mean it is for this. That is a distinction our communities have been teaching for a long time, and it feels like the machine world is only now arriving at the edge of understanding it.
The problem is two-sided, and I feel it from both sides at once. Communities should not have to choose between being excluded from AI's benefits and surrendering control over their knowledge in order to use AI-enabled services. And public institutions need practical rules — for classification, authority, permission, human oversight, procurement, vendor access, retention, and prohibited uses.
Right now, on most days, I am not sure anyone has those rules. And I am not sure how fast we can build them.
What Native Communities Are Worried About — and Why I'm Listening
These concerns are not hypothetical. They echo conversations I have had in community after community. Let me write them down the way people have shared them with me:
Extraction without consent. AI models are already trained on language materials, oral histories, songs, and cultural content that communities never authorized for that use. The pattern echoes older ones: knowledge taken and somehow valued more once it leaves the community's hands than it ever was while it stayed.
Cultural flattening and misrepresentation. Generative AI does not know what is sacred, what is seasonal, what is restricted, or what belongs only to certain families, clans, or ceremonies. Left unchecked, it produces generic, decontextualized, sometimes offensive, versions of living cultures. And those versions start showing up everywhere first.
Data colonialism, renewed. The same extractive logic is reaching tribal lands themselves, as data centers and AI infrastructure chase cheap energy and water. Communities are asking hard questions about who benefits from their land, their resources, their data. Again.
Loss of authority over knowledge. When community data enters AI systems, control is hard to get back. That is why frameworks like OCAP and the CARE Principles exist, and why tribes are investing in data governance as the next phase of sovereignty.
Language revitalization at risk. This one sits heavy with me. AI-generated language content can sound plausible even when it is wrong, flattening dialects, inventing forms, and threatening the authority of living speakers and knowledge holders. I grew up understanding that the language belongs to the people who speak it. A model that sounds right but is wrong is a different kind of loss.
The digital divide, widened. If AI-enabled services are built without Native communities at the table, we do not just miss out on benefits; we get governed by tools we never shaped. That is the fear I started with, and it keeps coming back.
These concerns shape every design choice I am exploring. Classification, AI-prohibited designations, human escalation, impact assessments, procurement clauses- this is not bureaucracy. It is our community's answers to these fears, made operational. That is the only way I can imagine doing this and sleeping at night.
An Idea I Keep Turning Over: The Indigenous AI Governance and Technical Assistance Commons
I have been developing a Native-led idea that pairs a practical Indigenous AI governance framework with secure, human-guided AI tools to expand access to organizational capacity-building and technical assistance. I call it the Commons, and it has two parts that have to work together:
A Model Indigenous AI Governance Standard. This would establish classifications: public, organizational, confidential, community-controlled, restricted, each triggering rules for authorization, permitted AI functions, human review, retention, secondary use, model training, vendor access, and removal. Restricted knowledge could be designated AI-prohibited: it simply does not enter AI systems at all. An Indigenous Knowledge and AI Impact Assessment and model procurement clauses would help institutions operationalize the standard, anchored in existing frameworks like NIST's AI Risk Management Framework.
A Commons of secure, human-guided tools. Approved-resource knowledge assistance; application and reporting support; needs assessment and triage; data and evaluation assistance; resource development tools; and a secure practitioner workspace. AI handles the appropriate navigation and preparation tasks, while complex or sensitive needs escalate to qualified human practitioners and subject-matter experts. Human judgment stays at the center.
Here is the part I keep reasoning out loud about. I do not want communities to have to choose between safety and usefulness. The whole reason for human-guided design is that I refuse to accept that trade. I want the tools to be genuinely helpful, not performatively cautious, not gatekept, and I want them built so the community stays in control of what enters them. Whether that is actually possible, I am still testing. I want to find out honestly, in public, rather than pretend I already know.
A Hawaiʻi Pilot That Can Be Adapted, Not Imposed
The Commons would begin with a Hawaiʻi policy and implementation pilot, bringing together state data and technology leadership, Indigenous data stewards, Native-serving organizations, privacy and procurement expertise, and technical staff.
Starting at home matters to me. I want to build this where I can be closest to the people it is for, where the accountability is not abstract. And I want the outcome to be a transferable architecture, something communities and governments adapt to their own authority and protocols.
One rule imposed across all Indigenous peoples is not the goal. That would just be a new kind of extraction. The model must be adaptable so that Tribal governments, Alaska Native communities, Native Hawaiian communities, and Indigenous peoples of U.S. territories can define their own permissions and restrictions. I do not claim to know what those should be for anyone but my own communities. I want the framework to leave room for them to decide.
Where This Leaves Me
I am writing all of this down because I am not done struggling with it. I want AI-enabled capacity building that is genuinely useful, and governed by rules that keep authority over Indigenous knowledge with the people and institutions entitled to exercise it. Communities should shape both the rules and the practical uses of civic AI, not just be governed by them.
I believe that. I also believe I do not have it fully worked out. Both are true, and I am trying to be honest about them.
This is the first installment of an ongoing series, a place to think out loud and test ideas in public. Future parts will go deeper into the governance standard, the classification framework, the pilot, and the specific concerns our communities have raised. If this resonates with the work you are doing- funders, policymakers, fellow practitioners, community leaders- I would welcome the conversation. I think we need each other to get this right.
— Olani Lilly, CoMission LLC · mission.consulting
