Māori Algorithmic Sovereignty

• Māori Algorithmic Sovereignty is a framework defining Māori control over algorithms using culturally grounded principles and legal standards.
• It mandates Māori participation across the entire algorithmic lifecycle, integrating indigenous customs with modern data and code governance.
• Its application yields measurable benefits, including reduced bias and enhanced transparency in systems impacting Māori communities.

Māori Algorithmic Sovereignty (MASov) defines the right of Māori to exercise control and governance over algorithmic systems that either utilize Māori data or are applied to Māori individuals, communities, or environments. Extending well-established principles of Māori Data Sovereignty (MDSov), MASov encompasses the entire algorithmic lifecycle, mandating Māori participation, authority, and oversight from system conception and design through deployment and lifecycle management. Its core objective is the meaningful protection and expression of Māori rights, values, and interests within emerging data-driven technologies, ensuring that algorithms operationalize and are accountable to Māori legal, cultural, and ethical frameworks (Brown et al., 2023).

1. Conceptual Foundations: From MDSov to MASov

Māori Data Sovereignty (MDSov) asserts that all data about Māori people, culture, language, and taonga, including digital forms, fall under Māori laws and governance based on Te Tiriti o Waitangi (the Treaty of Waitangi). The Treaty’s triad—Partnership (Article 1), Protection (Article 2), Participation (Article 3)—is implemented through six tikanga (customary law)-based principles: Rangatiratanga (Authority), Whakapapa (Relationships/Data lineage), Whanaungatanga (Obligations/Kinship), Kotahitanga (Collective Benefits), Manaakitanga (Reciprocity/Respect), and Kaitiakitanga (Guardianship).

MASov emerges to address the socio-technical complexity of algorithmic systems which involve not only technical procedures but also motives, funding, design, deployment, human judgment, and ongoing governance. Once an algorithm ingests Māori data or impacts Māori communities or taonga, the entire array of protection, participation, and partnership rights under MDSov extend to the complete algorithmic system.

2. MASov Principles and Sub-principles

Retaining the six primary MDSov principles, MASov contextualizes each to algorithmic application. Every principle decomposes into three sub-principles, specifying actionable governance requirements at all stages.

MASov Principle	Sub-principles (Algorithmic Context)
Rangatiratanga	1. Control: Māori control motives, inputs, core, and maintenance <br> 2. Jurisdiction: Data/code must reside within Māori jurisdiction <br> 3. Self-determination: Participation must enable sustainable self-governance
Whakapapa	1. Transparency: Full process and lifecycle transparency <br> 2. Data Relationship: Trace usage, obey MDSov throughout <br> 3. Sustainability: Inputs/outputs must yield durable Māori benefit
Whanaungatanga	1. Balancing Rights: Individual/collective rights, risks, benefits balanced <br> 2. Redress: Māori must have mechanisms to challenge decisions <br> 3. Accountability: Practitioners accountable to Māori affected
Kotahitanga	1. Benefit: Maximize benefits, minimize harms for Māori <br> 2. Capacity Building: Engagement at every stage to build capability <br> 3. Solidarity: Support connection with other Indigenous/affected groups
Manaakitanga	1. Respect: Uphold mana of individuals/communities <br> 2. Privacy: Safeguard privacy across data lifecycle <br> 3. Consent: Ensure free, prior, informed consent at all stages
Kaitiakitanga	1. Protection: Stewardship of all algorithmic components <br> 2. Ethics: Māori protocols and knowledge underpin use/access <br> 3. Restrictions: Māori determine tapu (restricted)/noa (open) status of elements

These principles apply from motivation and design, through data selection, modelling, deployment, to monitoring and redress.

3. Rationale, Cultural Logic, and Legal Foundations

MASov is anchored in the triad of Te Tiriti o Waitangi, mapping Partnership with Rangatiratanga and Manaakitanga, Protection with Kaitiakitanga and Whakapapa, and Participation with Whanaungatanga and Kotahitanga. MASov principles not only mirror but explicitly extend data-centric governance to algorithmic provenance, with obligations to uphold mana, enable reciprocal benefit, strengthen kinship, and enforce stewardship over not only data, but code, model architectures, and outputs.

Contextualizing these frameworks is a history of colonial breaches of trust, persistent risk of extractive or biased technological practice, and the requirement of tino rangatiratanga (absolute chiefly authority) as recognized in Treaty relationships. MASov therefore enriches Western algorithmic governance with uniquely Indigenous concepts of authority, obligation, and guardianship.

4. Methodology for Decolonising Algorithmic Systems

Brown et al. advance a Kaupapa Māori mixed-methods methodology for decolonising existing algorithms:

Phase 1: Quantitative Bias Assessment

Standard algorithmic fairness metrics such as demographic parity and equalized odds are employed to establish whether Māori experience systematic disadvantage in algorithmic outputs, referencing benchmarks including Grother et al. (2019) on face recognition and Bartley et al. (2021) for social media bias.

Phase 2: Tikanga-based Qualitative Audit

Algorithms are decomposed into: Process, Motives, Inputs, Core, and Outputs. For each segment, MASov-derived questions are generated per principle (see paper Appendix 2). Māori stakeholder wānanga (workshops) enable surfacing of colonial logics and collaborative remedy design.

Phase 3: Quantitative Re-evaluation

Post-redesign, the revised system is reassessed with bias metrics. Changes may encompass new governance protocols, dataset revisions, and explainable model modifications.

Example: Audit of Algorithmic Inputs

• Rangatiratanga: "What controls do Māori have to mark inputs tapu/noa?"
• Whakapapa: "Do these inputs serve the algorithm’s Māori-defined motive?"
• Whanaungatanga: "Who is accountable for protecting these Māori data?"
• Additional queries for Kotahitanga (benefit), Manaakitanga (respect), and Kaitiakitanga (protection) are similarly instantiated.

5. Formal Models and Process Guidelines for Indigenised Algorithms

To transition from retrofitting toward the construction of Indigenised Algorithms, MASov is embedded in formal models and pipelines.

MASov Compliance Score

Quantitative evaluation is encapsulated by a compliance index:

$$S = \sum_{k=1}^6 w_k s_k, \qquad s_k = \frac{1}{3} \sum_{i=1}^3 s_{k,i}$$

Here, $s_{k,i}\in [0,1]$ indicates stakeholder-rated compliance for each sub-principle; $w_k \geq 0$ are weights set by policy ($\sum w_k = 1$). Threshold $S_0$ (e.g., 0.80) denotes the minimum for MASov compliance.

Constrained Learning Objective

In predictive modelling:

$$\min_\theta \quad \mathbb{E}_{x,y\sim D}\bigl[L\bigl(f_\theta(x), y\bigr)\bigr] \quad \text{s.t.}\quad C_{Rangatiratanga}(\theta) \leq 0, \ldots, C_{Kaitiakitanga}(\theta) \leq 0$$

Each constraint is a quantitative proxy for the corresponding MASov requirement (e.g., transparency, privacy leakage bounds). Appropriate constrained risk minimisation techniques, such as Lagrangian relaxations, enforce these governance constraints.

Pipeline for Indigenised AI

Pseudocode for MASov-compliant AI development:

1. co_define_motives(kaumatua, tangata_whenua) 2. co_select_data(communities, ensure_free_prior_informed_consent) 3. co_design_model(architectures, ensure_explainability=True) 4. define_MASov_metrics() 5. train_model(data, objective=L + λ·MASov_penalties) 6. evaluate_model(bias_metrics, MASov_metrics) 7. refine_with_wānanga(communities, results) 8. deploy_with_Māori_governance_board() 9. monitor_forever(algorithm_performance, MASov_compliance)

Each stage involves codifying ownership, tikanga protocols, and redress rights in formal governance documents.

6. Illustrative Use Case: Government Algorithm Audit

Brown et al. provide a pilot audit of a government welfare-eligibility algorithm:

• The model utilized an undocumented proxy feature (“area deprivation index”) that significantly under-represented Māori well-being.
• No mechanism existed for contesting risk scores or for Māori to exert oversight on data or model aspects (tapu/noa regime absent).
• Following Māori-led wānanga, interventions included:
  • Replacing deprivation indices with Māori-governed community survey data.
  • Implementing a real-time “Redress API” for human review of algorithmic outcomes upon Māori request.
  • Formalizing Māori kaitiakitanga through a memorandum of understanding (MoU) governing data and code repositories on an onshore “Māori cloud.”
• Bias retesting after these interventions showed over a 40% reduction in disparate false-negative rates for Māori claimants.

This suggests that applying MASov methodologies can yield measurable reductions in algorithmic disparity and enhance both procedural and substantive Māori governance.

7. Towards Transformational Indigenised Algorithmics

Retrofitting MASov to extant algorithmic systems constitutes only a transitional solution. The ultimate objective is to design and deploy Indigenised Algorithms whose socio-technical logic, architecture, and governance instantiate Māori sovereignty (tino rangatiratanga) from inception. The envisioned MASov-compliant system is “nested” within a protective and enabling shell of Māori value systems, such that all algorithmic practices—from data ingestion to ongoing operation and redress—are intrinsically indigenous in method and output. This structural shift not only satisfies Western fairness standards, but realizes deeply Māori paradigms of justice, accountability, and collective benefit (Brown et al., 2023).