1 of 14

Local AI in Action: Global Examples of Indigenous Knowledge Language Models

Elizabeth F. McLean, Knowsaic

lizmclean@knowsaic.net

https://www.linkedin.com/in/efmclean /

KM Triversary Forum

Tuesday October 14, 2025 (virtual)

2 of 14

The Epistemic Divide

Problem: Western-centric Large Language Models (LLMs) amplify “epistemic injustices” and cultural homogenization

AI Paradox

Economic pressures for generalized models
Urgent need for “pluralism” with contextually grounded solutions

Localized, autonomous systems ARE actively being built

KM Triversary Forum 2025 Elizabeth F. McLean

2

3 of 14

Scholarly Mandates: Theoretical Foundation

KM Triversary Forum 2025 Elizabeth F McLean

3

ETHICAL REQUIREMENTS

TECHNICAL IMPERATIVES

DATA SCARCITY AND SOVEREIGNTY

Indigenous Data Sovereignty (IDS)
Reciprocity and interconnectedness
Cultural relevance
Consent, Credit, Compensation, Control

Low Resource Languages (LRL’s)

Synthetic data
Cross-lingual Transfer Learning

4 of 14

The Policy Arc (2001 → 2025)

KM Triversary Forum 2025 Elizabeth F McLean

4

🌐

2021–2025: Knowledge Sovereignty (KM4Dev, RealKM, Indigenous KM literature)

2017–2021: Pluralism & Ethical AI (UNESCO AI Ethics)

⚖️

2001–2007: Linguistic & Cultural Rights (UNESCO, UNDRIP)

🕊️

5 of 14

Five Inclusion Pillars

KM Triversary Forum 2025 Elizabeth F. McLean

5

🌐 Pluralism�🧠 Epistemic Inclusion�🤝 Co-Governance�🏛️ Institutionalization�⚖️ Ethics

6 of 14

The Practice Curve (2020 →2025)

Reality Bridge 1: AingA.I .

Indigenous Data Sovereignty (IDS)
Interpretation and translation
Localized operations instead of Silicon Valley
Audio and Text Capture
Incorporate all dialects

KM Triversary Forum 2025 Elizabeth F McLean

6

Emerging Models	Region/Local
Māori BERT	Oceania
SeaLLMs	SE Asia
Komodo-7B	Indonesia
Egune AI	Mongolia
Sahabat-AI	Indonesia
Latam-GPT	Latin America
AfriLLM YorubaBERT	Africa
LakotaBERT	US (South Dakota)
FLAIR AINGA.I Inuktituk	(Canada / global ASR)

Inuktitut and other Indigenous languages. This initiative serves as a model for upholding Indigenous data sovereignty in the development of artificial intelligence (AI).

The primary features and goals of the AingA.I. project include:

• Technology and Function: AingA.I. is developing an interpretation and translation app powered by AI. The goal is to make learning Inuktitut more accessible. The technology is planned to be rolled out as a plug-in to existing apps or websites such as YouTube.

• Target and Scope: The aim is for the future of AI in Inuktut to be "for Inuit, made by Inuit, for the benefit of all Inuit", moving away from benefiting external companies like those in Silicon Valley. The current work focuses on collecting text and audio from language experts in the South Baffin dialect, with a long-term plan to eventually incorporate all the different Inuktut dialects into the app. The app is intended to serve as an additional tool for learning, not a replacement for human interpreters and knowledge.

• Data Sovereignty and Ethics: A core guiding principle is ensuring Indigenous communities retain control and copyright over their linguistic data and the systems created. This approach directly operationalizes Indigenous data sovereignty. The project aligns with ethical principles that demand Consent, Credit, Compensation, and Control (the 4 Cs) when dealing with Indigenous data.

• Development Philosophy: The company's founder and CEO, Kirt Ejesiak, hopes other Indigenous companies will be inspired by their invention and potentially partner with them for other language revitalization efforts, stating, "We want to control our destiny". Their work is designed to develop repeatable solutions that Indigenous communities can realistically build themselves.

• Challenges: The founder noted the scarcity of Inuit representation in the AI field and emphasized the need for policymakers and governments to invest in and protect the intellectual property that belongs to Indigenous people.

7 of 14

The Practice Curve (2020 →2025)

Reality Bridge 2: These LLMs address Culture, Arts and Authentic Data

Egune AI – Mongolia

National AI Infrastructure
Cultural ways of living specializing in Nomadic traditions
Uses synthetic data vital LRL’s

Lelapa AI (South Africa)

Collaborates with linguists and local communities for authenticity

KM Triversary Forum 2025 Elizabeth F McLean

7

8 of 14

The Practice Curve (2020 →2025)

Reality Bridge 3: Technical Scale and Sovereignty

KM Triversary Forum 2025 Elizabeth F McLean

8

Egune AI – Mongolia

Use of synthetic data are vital strategies LRL’s

Sahabat -AI – Bahasa Indonesia

Open-Source ecosystem
Promotes Digital Sovereignty
Empowers cultural nuances national specific context

9 of 14

Impact Alignment with UN Global Goals

Indigenous Knowledge Systems are development accelerators

SDG 4
SDG 10
SDG 16
SDG 13
SDG 15 –Traditional Ecological Knowledge (TEK)

KM Triversary Forum 2025 Elizabeth F McLean

9

10 of 14

I’ll be monitoring these developments

KM Triversary Forum 2025 Elizabeth F McLean

10

🌐 Ontologies for Trustworthy Indigenous Knowledge AI (IK-AI)

🧠 Synthetic data for Low-Resource Languages (LRLs)

Tools for voice and other multi-modal assets like Automated Speech Recognition (ASR)

🏛️ Archival Partnerships

�

Ontologies for Trustworthy Indigenous Knowledge AI (IK-AI)

Ontologies are a cultural and ethical necessity—not just a technical requirement—for properly representing Indigenous worldviews within AI systems. They provide a structured framework to formally define the concepts and relationships of a knowledge domain.

�

Core Need: Why Ontologies are Essential for IK-AI

Ontologies address major ethical and technical issues when integrating Indigenous Knowledge (IK):

FunctionBenefit/RequirementSourcePreventing AppropriationStop cultural distortion, erasure, or "data colonialism" by properly representing complex, holistic knowledge.Building TrustGround "black box" LLMs in explicit, structured knowledge, allowing the AI to explain its conclusions in a human-understandable way.Ensuring SovereigntySupport Indigenous Data Sovereignty and intellectual property (IP) rights by requiring co-creation with free, prior, and informed consent.Addressing BiasCounteract systemic bias inherent in Western-centric datasets and accurately represent diverse Indigenous perspectives.Enabling Co-CreationTransform Indigenous communities into active stewards and creators of technology that aligns with community values.

�

Comparison: Western vs. Indigenous Ontologies

The fundamental difference lies in their approach to reality and knowledge generation.

AspectWestern-Centric Ontology (Standard AI)IK Systems OntologyNature of KnowledgeAbstract and data-driven.Relational, community-centered, and grounded in lived, holistic experience.Ethical ScopeOften focused on human subjects (implied).Captures complex relationships, including environmental ethics (e.g., Kaitiakitanga or guardianship).Development ProcessUsually siloed or top-down (implied).Requires a participatory development approach, engaging knowledge holders throughout the AI lifecycle.

Synthetic Data: Synthetic data is a novel and increasingly popular way to solve the fundamental problem of data scarcity, particularly when developing advanced AI systems like Large Language Models (LLMs) for Low-Resource Languages (LRLs).

1. Purpose (Resource Augmentation): Synthetic data serves as a resource augmentation method, filling the gaps where there is insufficient or scant real data available for training models. This approach enriches the training dataset, thereby improving the model’s learning efficiency and overall effectiveness, particularly in Natural Language Processing (NLP) tasks. This method also helps to conserve resources and mitigate the reliance on extensive annotated datasets. For example, the Egune AI system for the Mongolian language successfully used synthetic data to augment scarce Mongolian text, directly addressing the data scarcity challenge.

11 of 14

Conclusion:

Truly beneficial AI requires a structural shift toward:

Intentional Governance
Collective Action: cooperative models and community-led development to democratize access to scarce resources, such as shared computing infrastructure
Indigenous Data Sovereignty: inherent right of knowledge-holders to retain Control, Consent, Credit, and Compensation over their linguistic data and systems, thereby preventing new forms of exploitation or data colonialism

KM Triversary Forum 2025

11

This Photo by Unknown Author is licensed under CC BY-ND

The path toward realizing the potential of Large Language Models (LLMs) in Low-Resource Languages (LRLs) rests on resolving a fundamental tension: balancing the necessary pluralism of local, culturally grounded systems with the economic pressure toward generalized global models.

Achieving truly beneficial AI requires a structural shift toward:

1. Intentional Governance: We must move beyond reactive development toward intentionality and proactive ethics. This means integrating principles like relational accountability and intergenerational responsibility into the AI lifecycle.

2. Collective Action: Solutions must prioritize cooperative models and community-led development to democratize access to scarce resources, such as shared computing infrastructure.

3. Indigenous Data Sovereignty: This collective approach is anchored in upholding the inherent right of knowledge-holders to retain Control, Consent, Credit, and Compensation over their linguistic data and systems, thereby preventing new forms of exploitation or data colonialism

The core shift regarding Indigenous Knowledge (IK) systems is a transition from being passive subjects of technological scrutiny or data extraction to becoming active, sovereign agents who are co-creating and governing artificial intelligence (AI) systems according to their own ethical and cultural frameworks.

This shift is necessary because the development of Large Language Models (LLMs), traditionally dominated by high-resource languages and Western perspectives, has historically led to "epistemic injustices" and cultural homogenization, which threatens to erase cultural histories and diverse ways of thinking. The change is driven by a focus on three critical dimensions:

1. Shift from Extracted Data to Data Sovereignty

The most significant operational shift involves asserting and implementing Indigenous Data Sovereignty (IDS).

• From Extractive Logic to Control: The historical and technological model of "data colonialism"—where data and insights are extracted from Indigenous communities without consent or benefit-sharing—is being actively rejected

12 of 14

AI Use Statement

KM Triversary Forum 2025 Elizabeth McLean

12

NotebookLM was used to store and summarize sources I selected in support of my topic about examples of emerging indigenous large language models beyond an existing body of research and UN policy frameworks.

13 of 14

UNESCO History of Recommendations

KM Triversary Forum 2025

13

UNESCO. (2001). Universal Declaration on Cultural Diversity. Paris: UNESCO.

UNESCO. (2003). Convention for the Safeguarding of the Intangible Cultural Heritage. Paris: UNESCO.

UNESCO. (2005). Convention on the Protection and Promotion of the Diversity of Cultural Expressions. Paris: UNESCO.

United Nations. (2007). United Nations Declaration on the Rights of Indigenous Peoples (UNDRIP). New York: United Nations.

UNESCO. (2021). Recommendation on the Ethics of Artificial Intelligence. Paris: UNESCO.

14 of 14

Sources

KM Triversary Forum 2025

14

Indigenous knowledge | sustainable development goals—Resource centre. (n.d.). Retrieved October 19, 2025, from https://sdgresources.relx.com/indigenous-knowledge

Indosat Ooredoo Hutchison and GoTo Launch Sahabat-AI: Indonesia’s Open- Source LLM for Empowering Digital Sovereignty. (2024, November). GoTo. Retrieved October 19, 2025, from https://www.gotocompany.com/news

Maps—The world factbook. (n.d.). Retrieved October 19, 2025, from https://www.cia.gov/the-world-factbook/maps/

News, C. B. C. (2025, August 28). Inuit innovators turn to AI to revitalize Inuktitut. Eye on the Arctic. https://www.rcinet.ca/eye-on-the-arctic/2025/08/28/inuit-innovators-turn-to-ai-to-revitalize-inuktitut/

Sharma, R. (2025). Mongolian ai startup secures $3. 5 million to drive national ai infrastructure. Retrieved October 19, 2025, from https://www.thefastmode.com/investments-and-expansions/41648-mongolian-ai-startup-secures-3-5-million-to-drive-national-ai-infrastructure

Srreyansh, S. (2024). Advances and challenges in developing large language models for low-resource languages. Acta Scientific Computer Sciences, 6(5), 03–39. https://actascientific.com/ASCS/pdf/ASCS-06-0519.pdf

The Mongolian startup defying Big Tech with its own LLM. (2025, August 22). Rest of World. https://restofworld.org/2025/mongolia-egune-ai-llm/