The global AI conversation is focused on capability - what models can do, how fast, how accurately. But there is a prior question that most AI frameworks do not address: what should AI know about, and whose knowledge counts?
The Missing Layer
Current AI design follows a pattern. Collect data. Train a model. Optimise for accuracy. Deploy. Iterate. The framework is technical, and within its own terms, it works well. Models are getting better. Accuracy is improving. Enterprise applications are becoming viable.
But this framework carries assumptions that become visible when you examine them through an indigenous lens.
The assumption that knowledge is extractable - that it can be separated from its context, digitised, and processed by a system that has no relationship to its source. The assumption that more data produces better outcomes, regardless of the provenance or governance of that data. The assumption that optimisation for accuracy is sufficient, without asking accuracy for whom, and measured against whose understanding of truth.
These are not abstractions. They have practical consequences in every AI system deployed in contexts where indigenous communities are affected.
What Indigenous Knowledge Systems Offer
Indigenous knowledge systems - and I speak primarily from a Māori and Pasifika perspective, though these principles resonate across indigenous traditions globally - operate on fundamentally different premises.
Relational knowledge. In te ao Māori, knowledge is not an isolated object. It exists in relationship - to the people who hold it, to the land it comes from, to the generations who developed it, and to the purposes for which it is shared. Whakapapa (genealogy, interconnection) is not merely a cultural concept. It is an epistemological framework. Knowledge has lineage. That lineage matters.
Contextual authority. Not all knowledge is available to all people in all contexts. Some knowledge is held by specific communities. Some is shared only in specific circumstances. Some carries responsibilities that accompany access. This is not secrecy - it is governance. And it is governance that AI systems, which treat all accessible data as equally available for processing, do not understand.
Values-embedded design. When a Māori community builds a whare (house), the design encodes values. The orientation, the carvings, the materials - each carries meaning. The building is not separable from the values it embodies. This is a design principle that AI development has not yet learned: that systems encode values whether the designers intend it or not, and the question is whether those values are deliberate or accidental.
370M+
indigenous peoples worldwide, holding distinct knowledge systems that AI development has largely not engaged with
Source: United Nations, State of the World's Indigenous Peoples, 2023
Where This Matters in Practice
Health AI
Hauora (Māori health and wellbeing) is conceptualised through frameworks like Te Whare Tapa Whā - a model that encompasses taha wairua (spiritual), taha hinengaro (mental and emotional), taha tinana (physical), and taha whānau (family and social). An AI health system trained on Western biomedical data does not merely lack this perspective. It actively provides recommendations that may be culturally inappropriate or incomplete.
When we build AI for health contexts in Aotearoa, the question is not "how do we add a cultural layer?" The question is: "How do we design a system that understands health the way the communities it serves understand health?"
Education AI
Māori pedagogical approaches - ako (reciprocal learning), tuakana-teina (mentoring relationships), learning through whakapapa and narrative - are not alternative methods. They are established, effective educational practices with strong evidence bases. AI education tools trained on Western pedagogical assumptions risk reinforcing a single model of learning at the expense of approaches that serve Māori and Pasifika learners better.
Public Services
Government AI systems in Aotearoa operate within the context of Te Tiriti o Waitangi obligations. Data about Māori communities is subject to Māori data sovereignty principles articulated by Te Mana Raraunga. These are not optional considerations for public-sector AI. They are governance requirements.
Towards Cultural Intelligence
I use the term "cultural intelligence" deliberately. Not as a buzzword, but as a design requirement. AI systems that operate in culturally diverse contexts need the capacity to recognise, respect, and respond to different knowledge systems, not merely translate between them.
This is not about building "Māori AI" or "Pacific AI" as separate products. It is about building AI systems that are architecturally capable of holding multiple knowledge frameworks simultaneously, governed by the communities whose knowledge they engage with.
The technical challenges are real. But they are secondary to the governance and design challenges. We need:
- Participatory design that centres indigenous communities as co-designers, not consultants
- Data governance frameworks that honour indigenous data sovereignty
- Evaluation criteria that go beyond accuracy to include cultural appropriateness and community benefit
- Accountability mechanisms that give communities genuine authority over how their knowledge is used
An Invitation
This is not finished thinking. It is emerging thinking, informed by conversations with iwi, with researchers, with technologists, and with communities across Aotearoa and the Pacific. I share it not as a complete framework but as an invitation to engage with these questions seriously.
The AI industry is making design decisions right now that will shape how technology interacts with indigenous knowledge for decades. Those decisions should not be made without indigenous voices at the table.