When or If AI Meets Indigenous Knowledge?

An Ideation Series — Part 2: Why Classification Has to Come Before Innovation

In the first part of this series, I wrote about the question that keeps me up at night: who decides when, or whether, AI meets Indigenous knowledge. Since I put that out there, people have asked me, in the kindest way, what I would actually build. And the honest answer is that before I can build anything, I have to solve a problem that sounds boring but isn't: how do we sort what AI can touch from what it can't?

I keep using the word classification when I talk about this, and I can already feel people's eyes glaze over. Classification sounds like library work. But it's not about organizing things neatly. It's about deciding who has the authority to say, "This is ours, and this is not for the machine."

Here's what I mean. I've been thinking about categories like public, organizational, confidential, community-controlled, and restricted. Public is easy: things our communities have intentionally shared with the world. Organizational is trickier but familiar: the internal stuff an organization holds: files, plans, records, the way we work. Confidential information is information that would hurt people if leaked. Then there's community-controlled. This is the one that matters most to me and the one we talk about least. It's knowledge that belongs to a whole community, not to any one person or file, a story told only in certain seasons, a practice known only to certain families, an understanding of a place that a government record could never hold. And finally, restricted knowledge that should not move at all unless someone with the right to say so says so.

Now here's the part where I start to doubt my own categories. Because knowledge doesn't actually live in boxes. It lives in people. It moves through relationships, through ceremony, through seasons, through who you are and who your family is. I can name five categories on a whiteboard, but I know in my gut that a grandmother's recording doesn't care about my whiteboard. And that's the tension I keep walking into: if I want AI to be useful to our communities, I need clear rules about what it can and can't do. But the minute I write those rules down, I'm imposing a structure on knowledge that was never meant to fit one.

So what do I do with that? I don't think the answer is to throw out classification entirely. I think the answer is to admit that classification is only as good as the authority behind it, and that authority has to come from the community, not from me, and not from a tech company.

That's the part that scares me about the world rushing ahead. Governments and public systems are adopting AI at speed. And in that rush, somebody is going to classify our knowledge, I guarantee it. The only question is whether we're the ones doing the classifying, or whether it gets done to us by people who don't know what's sacred, what's seasonal, what's family, what's place. If we don't define these categories ourselves, someone else will define them for us, and their categories will be about their convenience, not our care.

I've also been wrestling with a harder question underneath all of this: who gets to make these calls? Is it the organization that holds a document? The speakers and knowledge holders? The elders? The families? The tribal government? The answer is probably all of the above, in different ways, for different kinds of knowledge, and that complexity is exactly why we can't just copy some generic data-privacy framework and call it done. Generic frameworks assume an individual owns their information. A lot of what our communities hold isn't owned by an individual. It's held in trust, for people who haven't been born yet.

That's the part I don't think the tech world has wrapped its head around. And I'm not sure I've fully wrapped my head around it either.

But here's what I do believe, even as I'm still working through the doubts: we have to get the classification conversation started now because innovation is already underway. The Commons I keep talking about, the Native-led initiative pairing governance with secure, human-guided AI tools, it can't work if we build the tools first and argue about boundaries later. Boundaries first. Then tools that respect those boundaries. That's the order. I know it's not the sexy order. But it's the respectful order, and I think respect is the whole point.

I want to build tools that genuinely help, that help a small staff write a grant at nine o'clock at night, that help a community show its impact without giving away its stories. But I refuse to build them on top of a system in which someone else decides what's public and what's sacred. If I can't get the classification right, I don't want the tools at all. That might sound like I'm being stubborn. Maybe I am. But I've watched too much of our knowledge be taken, so I'm casual about who holds the keys.

I don't have this solved. I want to be honest about that. What I have is a set of categories I'm testing, a deep belief that our communities have to be the ones defining them, and a willingness to be wrong and try again. In the next part of this series, I want to talk about the hardest category of all, the one that says some things should never meet AI at all. I'm still deciding what I think about that. But I think it's a question worth sitting with.

— Olani Lilly, CoMission LLC · mission.consulting