The Last Mile of Assimilation: Do Not Judge Our “Universe” by Your Logic

Introduction: In an Age of One-Click Images, Meaning Is Going Bankrupt

We live in an age exceptionally skilled at producing appearances and increasingly reluctant to contend with meaning. One click creates a poster; another produces a presentation; a third generates something that looks like research, a proposal, or an argument. “Output” is therefore no longer the scarce resource. What is scarce is the ability to distinguish whether an output has context, carries responsibility, and brings our understanding of the world any closer to the truth. This is already dangerous enough for mainstream society. For Indigenous Peoples, it may be more devastating still. When languages, ceremonies, restrictions, memories embedded in landscapes, and traditional knowledge are flattened into “material for AI to reproduce,” the result is not digital preservation. It is assimilation for a second time.

The question has never been only how fast a technology operates, but whose logic it uses to decide what counts as knowledge. Mainstream civilization is adept at presenting its own classifications as universal standards: if something can be translated, it exists; if it can be quantified, it is credible; if it can be converted into data, it can be governed. Yet for many Indigenous Peoples, language is not a label, restrictions are not cultural decoration, and cosmology is not an optional backdrop. These are norms, obligations, and practical orders through which communities sustain relationships with land, ancestral spirits, seasons, and one another. If AI cannot comprehend these dimensions but rushes to generate, classify, explain, and reuse them, it merely renders a mainstream illusion more fluently.

The Real Danger of AIGC Is Not Empty Content, but Content That Only Looks Substantive

Today’s AIGC resembles a visual fast-food outlet open around the clock. Ask for an epic tone and it supplies one; ask for an academic look and it supplies that; ask for a mysterious “Indigenous style” and, within seconds, it serves a whole table of material that appears rich in flavor. The problem is that such content often completes the task of looking right without meeting the obligation to be right. It appears to satisfy a demand while obscuring provenance, context, and responsibility. We receive an abundance of outputs shaped like answers and fewer tools for knowledge that can be applied, verified, and held accountable to conditions on the ground.

Many people treat this as a minor defect and call it hallucination, as though a model’s occasional fabrication were merely the growing pains of an emerging technology. In high-risk settings, hallucination is never a joke. During a disaster, when traditional knowledge is governed by restrictions and conditions of use, or when the interpretation of meaning bears on land ethics and community decisions, an error is not a charming imperfection. It is a systemic problem that transfers risk to the community. More troubling still, a mainstream model does not necessarily fabricate clumsily. It may sound like an expert—more fluent, even, than someone who genuinely understands the community context. I call an authoritative-sounding error generated from the standpoint of mainstream civilization a “mainstream hallucination.” It is more than ordinary information bias; it packages a mainstream worldview as fact itself.

Assimilation Runs Deepest Not in Vocabulary, but in the Foundations of Reasoning

Mainstream society often imagines language preservation as dictionaries, recordings, courses, subtitles, and databases. All are important, but by themselves they preserve only the surface. The more difficult task concerns the forms of reasoning carried by Indigenous languages and traditional knowledge. Some knowledge can be understood only in a particular season. The key to some expressions lies in metaphor, pragmatics, and context-dependent authority. Some restrictions are not superstition but boundaries that protect the safety of communities and environments. Adding this material to a Traditional Chinese corpus and attaching a few cultural labels will not automatically enable a model to understand it.

This is precisely why mainstream civilization so often underestimates the depth of assimilation. It assumes that translating a language, filming a ceremony, or organizing a narrative into a story completes the work of preservation. For Indigenous Peoples, however, what is lost is often not a handful of words but an entire structure for understanding the world, responding to risk, assigning responsibility, and sustaining relationships. If AI ultimately breaks these structures into searchable, displayable, and marketable fragments, it is not supporting cultural continuity. It is carrying assimilation through its last mile—more gently, more digitally, and more efficiently in appearance, but more comprehensively in effect.

Two-Eyed Seeing Is Not a Slogan; It Is a Necessary Corrective to Mainstream-Centered Thinking

This is why I continue to emphasize Two-Eyed Seeing. It does not mean treating Indigenous knowledge as a local embellishment to Western science, nor adding a few agreeable sentences about cultural respect to a report. Its real work is to recognize the strengths, limits, and appropriate contexts of different knowledge systems and allow them to operate in complementarity rather than letting a mainstream framework absorb the other. That work becomes especially important as climate change and disaster risks intensify. Reading landscapes, recognizing seasonal signs, observing ecology, maintaining oral protocols, and sustaining relationships with land are not romantic cultural relics. They are bodies of practical knowledge that may bear directly on survival.

For years, the IPCC and UNDRR have repeatedly recognized the critical value of Indigenous knowledge and local knowledge in climate adaptation and disaster-risk reduction. These international documents do more than attach a polite label saying that Indigenous Peoples “also contribute.” They make a substantive point: modern governance that trusts only a narrow set of mainstream indicators will miss extensive experience in resilience already operating on the ground. If AI merely produces more attractive reports or marketing copy, it remains an entertainment factory. If it can help turn Indigenous landscape memory, disaster-preparedness experience, seasonal patterns, and community collaboration into systems that can be passed on, taught, and used in scenario analysis, it may become part of resilience infrastructure.

Indigenous Data Sovereignty Is Not Optional; It Is the Baseline

All of these questions ultimately return to governance. Who defines the data? Who decides what may be made public? Who has authority to annotate it? Who may use it to train a model? If the answer is always an outside institution, platform company, research organization, or policy implementer, Indigenous knowledge can readily be extracted, rewritten, and redistributed in the name of “collaboration,” “preservation,” or “innovation.” The community is left with one role: providing content without controlling the rules. That is not collaboration. It is a cleaner form of mining for the digital age.

For that reason, Indigenous AI must begin not with technology deployment but with data sovereignty. The United Nations Declaration on the Rights of Indigenous Peoples affirms the rights of Indigenous Peoples to maintain, control, protect, and develop their cultural heritage, traditional knowledge, and traditional cultural expressions. The CARE Principles remind us that data governance is not only about access, but also about Collective Benefit, Authority to Control, Responsibility, and Ethics. These are not administrative appendices; they are the institutional floor. Without that foundation, any initiative claiming to “build AI for an Indigenous community” may look advanced while merely accelerating extraction for the mainstream world.

An Indigenous AI That People Can Actually Use Needs at Least Five Things

First, the community must define the task. A project should not begin with a model and then search for an Indigenous community to serve as its use case. AI should respond to community priorities, such as disaster warning, landscape interpretation, experience in farming, forestry, fishing, and animal husbandry, language transmission, Indigenous traditional medicine, or tiered governance for knowledge and restrictions associated with dream divination—not force a topic into existence merely to display technological sophistication. Second, knowledge must be placed under tiered permissions. Not all data are suitable for public release, and not all knowledge should be converted into a model. Some may be used only by community members, some only in a specific context, and some should not be digitized at all. Third, generations must be verifiable, with source attribution, traceable evidence, and retrieval-augmented mechanisms that work together against mainstream hallucination. Fourth, evaluation cannot rely only on conventional performance metrics; it must also test whether pragmatic meaning is correct, restrictions have been violated, and context-dependent authority has been respected. Fifth, governance must continue over the long term. A project cannot end with the data removed and the system becoming the property of an outside institution.

Stop Using Your Logic to Judge Our “Universe”

What I most want to emphasize is that mainstream society repeatedly treats intelligibility to the mainstream as a condition for knowledge to exist. This is assimilation’s deepest violence. Your logic may be one tool, but it cannot be the final judge. The Indigenous universe is not a repository waiting to be translated into convenient products and public teaching materials. It is a living worldview, system of responsibility, and order of knowledge. When a mainstream AI model rearranges that universe according to its own grammar, it often imagines that it is broadening access when it is actually clearing the ground: deleting what it deems illogical, softening what it finds difficult to manage, and describing boundaries that should not be opened as insufficiently transparent.

That is why we must pause before AI today—not because we oppose technology, but because we refuse to let technology inherit the arrogance of mainstream civilization by default. AI may indeed be a final opportunity, but only if Indigenous Peoples participate in building it and meaning, rights, restrictions, evidence, and responsibility are designed into the system together. Otherwise, even the most capable model is merely another, more efficient machine of assimilation.

Conclusion

We can, of course, continue to generate attractive images, rapid proposals, and plausible arguments. But if these outputs do not make communities safer, knowledge more respected, and governance more traceable, they are ultimately only glitter on a keyboard. The goal worth pursuing is not to make the world look more like an advertisement, but to make it more understandable and more repairable. For Indigenous Peoples, this is not merely a choice about technology; it is a choice about civilization. Stop using your logic to judge our universe. Begin by recognizing that the universe contains more than one logic—and that this is precisely what AI may still have time to relearn.

Sources retained from the Chinese original

• United Nations | United Nations Declaration on the Rights of Indigenous Peoples (UNDRIP)
• Global Indigenous Data Alliance | CARE Principles for Indigenous Data Governance
• Intergovernmental Panel on Climate Change | Sixth Assessment Report and United Nations Office for Disaster Risk Reduction | What Is the Sendai Framework?
• Google Scholar | Two-Eyed Seeing and International Journal of Qualitative Methods | Using Two-Eyed Seeing in Research With Indigenous People