Data Is Not a Specimen: Indigenous Knowledge Cannot Be Locked in Someone Else's Drawer

Acknowledging Knowledge Does Not Equal Returning Power

In recent years, international discussions on climate change and biodiversity have increasingly referenced Indigenous knowledge. Reports state that Indigenous peoples and local communities have long cared for forests, wetlands, grasslands, coastlines, and agricultural landscapes; studies note that many land management practices are built upon careful observation, intergenerational memory, and community norms. These statements represent significant progress compared to past language describing Indigenous knowledge as "backward, superstitious, or undeveloped," yet this progress does not equate to full respect.

The most critical question has never been whether Indigenous knowledge is useful, but who holds the authority to decide how it is used.

Many institutions treat Indigenous knowledge like a very skilled presentation collector: they first praise it, then categorize it, and finally place it in their own drawers. Land memory becomes data fields; plant knowledge becomes datasets; hunting regulations become governance case studies; seasonal observations become climate adaptation materials. When cited, this is glorious; when used for decision-making, Indigenous voices are often absent. This is why data sovereignty matters: not to oppose data itself, but to refuse letting data detach from relationships and become someone else's asset.

Acknowledging the value of knowledge without simultaneously acknowledging governance rights often results in more refined extraction. In the past, some took land; now some take maps. Previously, specimens were removed; now data is taken away. Formerly, museums named civilizations; today, databases name knowledge. The interface has changed, but power relations have not fully shifted. This is where vigilance is most needed in the AI era.

Data Is Not Rainwater; It Is Drawn From Life

Modern policy loves to say "evidence-based." That phrase itself is correct; the problem lies in who defines evidence. If evidence only recognizes tables, satellites, surveys, models, and English-language journals, much Indigenous knowledge will be forced into institutional clothing that can be understood. More troublingly, once knowledge is converted into data, it may be copied, circulated, re-analyzed, commercialized, or even used for purposes the community did not consent to.

Data is not rainwater; it does not fall from the sky for everyone to freely access. Data is often drawn from certain people's languages, bodies, lands, labor, memories, and trust. Pumping water requires asking about water rights; extracting knowledge should also ask about data rights. Who collects? Who stores? Who may view? Who may modify? Who may delete? Who benefits? Who bears risk? If these questions are answered by external institutions, so-called collaboration is merely extraction wearing a polite coat.

Within the framework of Indigenous data sovereignty, data is not simply an information object; it is connected to people, land, relationships, responsibilities, language, history, and future. A certain plant name is not just a plant name; a route is not just coordinates; a legend is not just text; a disaster memory is not just a case study. They may involve harvesting rights, ancestral lands, family memories, seasonal taboos, public safety, ethnic identity, and community internal norms. Breaking these down into fields is easy; putting them back into relationships is difficult.

Climate Resilience Requires Relationships, Not Just Case Studies

Recent research again reminds us that Indigenous peoples and community lands play important roles in protecting carbon, biodiversity, and landscape resilience; yet these outcomes do not occur naturally, nor can they be replicated by simply applying a certain "traditional method" to a project. Behind them often lie taboos, divisions of labor, seasonal timing, patrols, sharing, responsibilities, elder judgment, and land ethics. Breaking these down into single technologies is like keeping only the drumbeat from a song and then claiming you have preserved music.

Climate policy's most common mistake is to recall Indigenous knowledge after disasters intensify, expecting it to quickly become scalable solutions. However, what truly resists climate risk is often not a single technique but an entire community governance capacity. Who may enter the forest? When can harvesting occur? How are water sources cared for? Which places should remain undisclosed? During abnormal weather, who notifies elders, schools, farmland, and roads? These are all resilience elements that do not necessarily resemble modern engineering.

For Taiwan, climate resilience cannot rely solely on central push notifications nor only on local temporary overtime. Indigenous township roads, streams, farmlands, community centers, long-term care services, school closures, and medical transportation often suffer together in a single torrential rain event. If AI systems view only rainfall grid points and map colors without local bridge memories, mountain road experience, elder care lists, and Indigenous language communication methods, they will see quickly but not necessarily accurately. Resilience is not merely information arrival; it is knowing what to do after information arrives.

Two-Eyed Seeing Is Not Helping Science Find Folklore Examples

Two-Eyed Seeing is often translated as "seeing with two eyes," but the emphasis is not on science using one eye to observe and another to decorate culture. True Two-Eyed Seeing means different knowledge systems hold each other accountable under shared goals. Scientific models can provide trends, imagery, risk estimation, and cross-regional comparisons; local knowledge can offer micro-topography, historical disaster memories, species behavior, community decision-making, and boundaries that should not be disclosed. When collaborating, it is not about one validating the other but about mutual responsibility.

In the AI era, this issue becomes sharper. Once data enters models, tracking its flow grows difficult. Indigenous languages, plants, place names, legends, rituals, routes, disaster memories, and care knowledge may all be "organized" into seemingly neutral training data. Models will speak fluently yet may not know which content should remain silent; they can combine beautifully but may not recognize which combinations are offensive. The most dangerous aspect of AI is not that it does not know boundaries, but that it does not realize what it does not know.

This also means cultural data governance cannot rely solely on a statement like "already anonymized." Many Indigenous datasets, even without personal names, can be re-identified through place names, vocabulary, plants, routes, mountain shapes, ritual timing, and community size. Moreover, harm may stem not only from privacy breaches but also from misinterpretation, commercial appropriation, tourism consumption, academic monopoly, and exposure of internal taboos. If data protection is understood only as individual privacy, collective rights are overlooked.

Sovereignty Is Not Locking Doors; It Is Deciding Which Door to Open

Data sovereignty does not mean sealing all data nor refusing research. Many communities need data to secure land rights, public services, health resources, educational support, climate adaptation, and cultural revitalization. The issue is that data governance must return to community self-determination: what can be publicly shared remains open; what should be shared stays within the community; restrictions remain where needed; withdrawal rights are honored. Data may flow, but it needs channels; knowledge may collaborate, yet collaboration requires owners.

Yuan Media AI discusses this not to turn "data sovereignty" into a pretty slogan, but to remind every person doing AI, research, policy, or media: Indigenous knowledge is not a free material library. It does not wait to be labeled, translated, uploaded, or fine-tuned. It has context, responsibilities, relationships, and the right to refuse.

If future climate policies truly need Indigenous knowledge, communities cannot merely be invited for conference photos; if future AI genuinely serves diverse knowledge systems, cultural boundaries cannot be treated as data cleaning inconveniences. Data is not a specimen. Specimens sit in drawers without voice; knowledge lives in relationships and will demand accountability. If civilization only wants the former and avoids the latter, that is not innovation—it is simply swapping old colonialism for new interfaces.

Start With Governance Processes, Not Data Downloads

If researchers, governments, or AI teams truly wish to collaborate with Indigenous communities, the first step should not be demanding data but co-designing governance processes. Who speaks for the community? Who may authorize? Where is data stored? How are outcomes returned? If model outputs err, who corrects them? Should a community later decide to withdraw, how are data and derivatives handled? These seemingly troublesome questions form the basic skeleton of lasting collaboration.

Practically speaking, cultural data projects should retain multiple tiers: public tier, community internal tier, restricted research tier, and non-digitizable tier. Not all knowledge should be photographed, recorded, transcribed, or vectorized. AI excels at diffusion; cultural governance requires boundaries. Without institutional connections between them, a disaster emerges: the better the tools, the faster inappropriate data spreads. That is not digital preservation—it is accelerated loss of control.