---
title: Decolonial Mindset Stack (DMS)
url: https://www.emergentmind.com/topics/decolonial-mindset-stack-dms
type: topic
---

# Decolonial Mindset Stack (DMS)

The Decolonial Mindset Stack (DMS) is a layered conceptual architecture for decolonial transformation in computing and adjacent technical fields. In its explicit named form, it was introduced in computing education as a seven-layer, non-linear framework that guides educators from foundational awareness of Indigenous exclusion toward active support for Indigenous sovereignty in computing education, grounding this progression in Freirean critical pedagogy, Indigenous methodologies, and relational lenses such as “About Me,” “Between Us,” and “By Us” [2509.18509]. Related literature has used the same label, or closely allied layered constructions, to organize decolonial practice in collaborative AI research, sociotechnical foresight in AI, NLP, and African data ethics, thereby extending DMS from educator transformation to data governance, participatory design, infrastructure, and institutional accountability [2206.03212], [2007.04068], [2311.08391], [2502.16043].

## 1. Genealogy and field-specific variants

The most explicit formulation of DMS appears in "Developing a Decolonial Mindset for Indigenising Computing Education (CE)" [2509.18509]. That work defines DMS as a seven-layered, non-linear framework for educator transformation: **Recognition, Reflection, Reframing, Reembedding, Reciprocity, Reclamation, and Resurgence**. Its stated purpose is to move computing educators from “mere awareness of Indigenous exclusion” to “active support for Indigenous sovereignty in computing education.”

Other works organize decolonial technical practice through comparable stacks. In collaborative AI research, DMS is presented as a five-layer structure comprising **Historical–Political Economy Foundation**, **Dependency & Power Dynamics**, **Data Empowerment Principles**, **Decolonial AI Methodologies**, and **Investigative & Reflexive Checklist** [2206.03212]. In Decolonial AI, the stack is articulated as three interlocking tactics: **Critical Technical Practice of AI**, **Reciprocal Engagements & Reverse Tutelage**, and **Renewal of Affective and Political Communities** [2007.04068]. In NLP, a DMS is proposed through the layered analysis of **data**, **algorithms**, and **software**, with interventions targeted at each material stratum [2311.08391]. In African data ethics, a six-layer architecture embeds the principles **Decolonize**, **Center All Communities**, **Uphold Universal Good**, **Communalism in Practice**, **Data Self-Determination**, and **Invest in Data Institutions & Infrastructures** into the data-science workflow [2502.16043].

| Context | Stack structure | Representative emphasis |
|---|---|---|
| Computing education | Seven layers | Educator transformation and Indigenous sovereignty |
| Collaborative AI research | Five layers | Dependency, empowerment, reflexive accountability |
| Decolonial AI | Three layers | Foresight, reverse tutelage, affective communities |
| NLP | Data, algorithms, software | Material coloniality in foundational artifacts |
| African data ethics | Six layers | Self-determination, communalism, institutions |

Taken together, these formulations suggest a family of decolonial architectures rather than a single canonical schema. What remains stable across them is the claim that coloniality is embedded not only in discourse but also in curriculum, funding, authorship, data infrastructures, modeling incentives, deployment channels, and governance arrangements.

## 2. Theoretical commitments and epistemic foundations

In computing education, DMS is explicitly grounded in **Freirean critical pedagogy** and **Indigenous methodologies**, especially **relationality**, **sovereignty**, **Two-Eyed Seeing**, and the **8-Ways of learning** [2509.18509]. Freire’s “conscientisation” underwrites Recognition; the reflection/praxis cycle underwrites Reflection; critique of the “banking model” underwrites Reframing; and praxis as enacted transformation underwrites Reembedding and the later layers. Indigenous methodologies supply the insistence that knowledge is situated, relational, reciprocal, and inseparable from community authority.

In Decolonial AI, the foundational vocabulary is the **coloniality of power**, together with **structural decolonisation**, **pluriversal epistemology**, and **reflexive power analysis** [2007.04068]. That formulation treats decoloniality as a demand to undo colonial mechanisms of power, economics, language, culture, and thinking; to reject the universalist assumption that only Western scientific culture generates valid knowledge; and to trace how values and norms become encoded into AI objects and institutions. The paper characterizes this as sociotechnical foresight aimed at centring vulnerable peoples and aligning AI research with beneficence and justice.

The collaborative AI formulation adds a political-economy substrate. It examines how colonial and neo-colonial histories shape present-day academic structures, R&D capacity, funding, authorship, and data control, drawing on concepts such as **neo-colonialism**, **dependence on the media of ideas**, and **data colonialism** [2206.03212]. The resulting stack asks who sets the research agenda, who controls resources, who can access and interpret data, and whether AI methods are selected to reduce rather than intensify dependence.

The NLP formulation contributes a **material lens on coloniality** using **Actor-Network Theory (ANT)** [2311.08391]. In that account, coloniality is not confined to representational harm; it is materially accumulated in internet infrastructure, script support, keyboard layouts, encoding standards, crowdwork pipelines, benchmarks, GPUs and TPUs, modeling frameworks, APIs, and deployed applications. The relevant actors include language speakers, content creators, curators, annotators, model developers, deployers, end users, policymakers, and a wide range of non-human technical artifacts.

African data ethics provides a normative articulation centered on six major principles: **Challenge Power Asymmetries**, **Assert Data Self-Determination**, **Invest in Local Data Institutions & Infrastructures**, **Utilize Communalist Practices**, **Center Communities on the Margins**, and **Uphold Common Good** [2502.16043]. The DMS derived from that framework treats these principles as interfaces in a layered workflow, each linked to concrete metrics and feedback loops.

## 3. The seven-layer computing education stack

The computing education DMS is the most granular layer-by-layer specification in the literature [2509.18509]. Each layer combines a definition, a relational lens, theoretical commitments, and practical activities.

**Recognition** is paired with the relational lens **“About Me.”** It is defined as developing foundational awareness of First Peoples’ histories, cultures, contributions to technology, and the ways colonial structures have excluded Indigenous voices from computing education. Practices include curriculum audits, storytelling sessions with Indigenous knowledge holders, and readings on colonial impacts in computing education. An illustrative formalization sets an educator’s self-assessed depth of awareness as $R_1 \in [0,1]$.

**Reflection** is paired with **“Of Us.”** It denotes critical self-examination of personal biases, privileges, and the subtle ways one’s teaching perpetuates colonial logic. Practices include guided journaling and facilitated Yarning Circles for collective critical dialogue. Its progression from Recognition is modeled as
$$
R_2 = f(R_1) = R_1 + \beta \cdot (1 - R_1), \quad 0<\beta<1,
$$
where $\beta$ captures engagement intensity in reflective practices.

**Reframing** is paired with **“With Us.”** It involves actively dismantling colonial logic embedded in curriculum design and pedagogy so as to create space for Indigenous worldviews. It draws on critique of the banking model and on Two-Eyed Seeing. Practices include syllabus “surgeries,” bias-spotting code reviews, and co-consultation with Elders and Indigenous scholars. Its conceptual progression is given by
$$
W_{i+1} = W_i + \gamma \cdot R_2, \quad \gamma>0.
$$

**Reembedding** is paired with **“For Us.”** It concerns co-creating and embedding Indigenous content, pedagogies, and protocols into computing curricula to ensure cultural relevance and community benefit. Practices include design jams with community partners, on-Country modules linking computing concepts to land stewardship, and new course modules on Indigenous technologies. Cumulative progress to this stage is summarized by
$$
S = \sum_{i=1}^{4} \alpha_i \cdot R_i, \quad \alpha_i>0, \quad \sum \alpha_i = 1.
$$

**Reciprocity** is paired with **“Between Us.”** It is defined as sustained relational accountability, with partnerships that flow knowledge, resources, and authority in both directions between educators and Indigenous communities. Examples include formalized MOUs, co-mentored research, community-driven evaluation of curricula, and iterative feedback mechanisms.

**Reclamation** is paired with **“Like Us.”** It is defined as respectfully emulating Indigenous ways of knowing, being, and doing, under the guidance of Elders and Indigenous scholars, thereby ethically enriching computing education. Practices include Elder-led workshops, integration of Indigenous languages and examples in code, and cultural protocols embedded in project workflows.

**Resurgence** is paired with **“By Us.”** It is defined as envisioning and supporting Indigenous self-determination in digital spaces through community-led design and governance of data, infrastructure, and algorithmic systems. Practices include community-owned platforms, Indigenous data governance policies, and support for Indigenous-led tech enterprises and research centers. A stated validation indicator is the number of Indigenous-led technology initiatives, leadership roles, and data-sovereignty frameworks enacted.

This seven-layer formulation reframes indigenisation as “not an endpoint but as a sustained ethical commitment to transformative justice and the co-creation of computing education with First Peoples” [2509.18509].

## 4. Non-linearity, composition, and formal models of progression

A defining feature of DMS is that layered order does not imply rigid linearity. The computing education paper states that the framework is “explicitly non-linear and iterative” [2509.18509]. Earlier growth in Recognition, Reflection, and Reframing supplies the awareness and collaborative orientation needed for Reembedding, Reciprocity, Reclamation, and Resurgence; however, educators may loop back from Reembedding to Reflection as new community insights surface, move directly from Reflection to initial Reciprocity by establishing community dialogue, or advance to Reclamation once reciprocal trust affords mentorship relationships.

That interaction is formalized as a dynamic system:
$$
R_{i+1} = R_i + \phi_i \cdot (\mathrm{Capacity}_i - R_i), \quad \phi_i \in (0,1), \quad i=1\ldots 6,
$$
where $\mathrm{Capacity}_i$ is the educator’s relational readiness to move into the next layer. Over time, the vector $R=(R_1,\ldots,R_7)$ is described as converging toward sustained engagement across all layers.

The Decolonial AI formulation expresses interdependence differently but with the same compositional logic:
$$
\mathrm{DMS} = \mathrm{Layer}_3\Bigl(\mathrm{Layer}_2(\mathrm{Layer}_1(\mathrm{AI\ Practice}))\Bigr).
$$
In that schema, reflexive technical practice yields artifacts such as model documentation and contextual fairness analyses; reciprocal engagements use those artifacts in co-design and dialogue; and affective and political communities provide the solidarity and institutional durability needed for ongoing reflexivity and exchange [2007.04068]. The paper explicitly states that a failure at any layer yields brittle or colonial AI.

The collaborative AI and African data ethics versions also rely on feedback architectures. In collaborative AI, the top-layer reflexive checklist is applied at project inception, mid-term review, and post-mortem, and its responses are tied to action items in a reflexivity matrix [2206.03212]. In African data ethics, every adjacent layer exchanges signals: “Decolonize” emits a coloniality risk score; “Center All Communities” requires a non-zero engagement vector for each community; “Uphold Universal Good” issues a go/no-go; “Communalism in Practice” manages consensus and restorative responses; “Data Self-Determination” enforces ownership and consent; and “Invest in Data Institutions & Infrastructures” publishes an annual readiness report that resets the stack for the next cycle [2502.16043].

A plausible implication is that DMS is best understood as a recursive governance pattern: layered commitments are sustained through repeated reassessment rather than through one-time compliance.

## 5. Operationalization across computing, AI, NLP, and data science

The computing education literature provides concrete enactments of DMS. For Layers 1 and 2 it cites curriculum audits, Yarning Circles at the University of Canberra, and guided journaling in staff workshops. For Layer 3 it cites syllabus surgery and bias-spotting code reviews at Deakin University in consultation with First Nations Elders. For Layer 4 it cites co-created design jams producing on-Country coding modules, including modules connecting GIS concepts to Indigenous land stewardship. For Layer 5 it cites formal MOUs for co-mentored research, including Deakin Cyber’s Indigenous Storytelling Project on Cybersecurity. For Layer 6 it cites Elder-led cultural workshops and integration of Indigenous languages into programming assignments. For Layer 7 it cites support for community-owned digital archives and Indigenous data governance frameworks guided by AIATSIS ethics [2509.18509].

Assessment mechanisms in that same work include portfolio-based reflections, community feedback loops evaluating relational accountability, and the **Transformer Index**:
$$
T = \frac{(|\mathrm{Partnerships}| + |\mathrm{Indigenous\text{-}Led\ Initiatives}| + \mathrm{EngagementDepth})}{\mathrm{NormalizationFactor}}.
$$
The paper also recommends mixed methods: self-reflection surveys, community feedback, and quantifiable indicators of Indigenous leadership in computing education.

In collaborative AI research, operationalization begins with political-economy audit and dependency analysis. Proposed steps include collecting and comparing R&D spending data in PPP, mapping researcher density and infrastructure gaps, auditing grant sources and PI or Co-I affiliations, tracking data collection versus analysis roles by geography, negotiating equitable authorship and budget transparency, ensuring multilingual low-barrier open-access data platforms, investing in local training and technology transfer, and designing data architectures for local sovereignty [2206.03212]. The AirQo case is used to instantiate these layers through Google AI Impact Grant and EPSRC funding, Makerere-led PI structure, a public app providing free access to calibrated air-quality data, technical mentorship, and Gaussian Process calibration of low-cost sensors.

In Decolonial AI, operational practice is embedded in everyday R&D through workshops on decolonial theory, project-specific DecolonialConstraints, layered sprints, quarterly “Friendship Audits,” and rotating oversight by community representatives [2007.04068]. Layer 1 activities include multi-dimensional value checkpoints, domain translations of technical fairness, Safety-in-Context reviews, and policy design with local institutions. Layer 2 includes intercultural dialogue platforms, documentation reciprocity, and participatory action research. Layer 3 includes support for grassroots AI collectives such as Data 4 Black Lives, Black in AI, Queer in AI, and Deep Learning Indaba.

The NLP literature operationalizes decoloniality at three material strata [2311.08391]. At the data layer, interventions include funding offline and low-bandwidth corpora creation under local governance, providing native-script keyboard and encoding tools, adopting expanded Data Statements co-authored by affected communities, using linguistically informed sampling, and ensuring free, prior, informed consent. At the algorithm layer, it recommends few-shot or parameter-efficient tuning, pruning and distillation for local hardware compatibility, community-defined success metrics, and transparent reporting of compute, carbon, and provenance. At the software layer, it recommends FLOPs-efficient inference code geared to consumer hardware, multilingual open-source toolkits, co-designed deployments, community advisory boards, and artifact reporting in local languages.

African data ethics extends operationalization with formal layerwise criteria [2502.16043]. These include **ColonialityIndex**, **EngagementCoverage**, **NetBenefitRatio**, **MinDignityViolation**, **ConsensusRate**, **RestorativeClosureTime**, **SovereigntyCompliance**, **IndigenousKnowledgeIndex**, **CapacityGain**, and **EducationUptake**. The purpose is to bind decolonial aspiration to measurable workflow obligations rather than leaving it at the level of general principle.

## 6. Reframing exclusion, sovereignty, and ongoing ethical commitment

A central contribution of DMS is the reframing of familiar technical problems as manifestations of coloniality rather than as isolated deficits. In computing education, DMS explicitly shifts discourse on “underrepresentation” from an individual-deficit model to one of **systemic exclusion rooted in colonial legacies** [2509.18509]. Under this account, underrepresentation is not a mere recruitment problem; it is a call to interrupt the systemic cycle of marginalisation through relational engagement, curricular redesign, and community-led resurgence.

In collaborative AI research, the corresponding reframing concerns **dependency**. The stack asks who funds the work, who sets the agenda, who occupies PI roles, what authorship prospects exist for indigenous researchers, whether outputs are open-source or open-access, how ongoing costs are minimized, and how empowerment is measured [2206.03212]. The AirQo example makes this concrete by contrasting UK and Uganda R&D capacity and by foregrounding questions of self-sustainability and local control.

In Decolonial AI, the problem is articulated as the persistence of colonial mechanisms in how AI is built, governed, and deployed [2007.04068]. The paper explicitly states that decolonial AI does not seek to overthrow political colonialism today; rather, it seeks to dismantle persistent structures in contemporary technical practice. This is linked to the idea of “algorithmic coloniality,” where values and power asymmetries are invisibly encoded into datasets, models, and institutions.

The NLP literature adds a material caution: combating coloniality requires “not only changing current values but also active work to remove the accumulation of colonial ideals in our foundational data and algorithms” [2311.08391]. Empirically, that work reports that inequality along colonial boundaries increases as NLP builds on itself, with 60.3% of LREC papers from Europe alone in the annotation phase, China accounting for 29.4% of “our-model” papers in the modeling phase, over 50% of GPU-using papers authored from the USA and China, zero deployment papers from Middle East and North Africa, Sub-Saharan Africa, and South America in the sampled industry and in-practice tracks, and Shannon Equitability and Gini remaining in the “extreme inequality” band even as paper volume grows.

African data ethics pushes the same argument toward institutional design: pluralism in global data ethics is operational only if it challenges power asymmetries, asserts self-determination, utilizes communalist practices, centers communities on the margins, and invests in local institutions and infrastructures [2502.16043]. This suggests that DMS is not exhausted by representational inclusion. Its more demanding endpoint is shared authority over knowledge, data, infrastructure, and the terms of technical futures.

## 7. Conditions for implementation and sustainability

The computing education formulation identifies several conditions for durable implementation: **institutional investment** in professional development and community partnerships; **recognition of decolonial work in promotion criteria**; **relational readiness** among educators and leaders; **non-linear pathways** revisited according to local context and community priorities; **ethical governance** through codes such as AIATSIS, consent, data sovereignty, and compensation for community reviewers; **measurement and reflection** through mixed methods; and **sustained collaboration** embedded in ongoing curriculum cycles, research agendas, and technology governance structures [2509.18509].

Collaborative AI research adds a self-sustainability criterion. Its reflexive checklist includes questions not only about inclusion and authorship but also about whether local actors are meaningfully integrated, whether outputs will remain open and usable, and how post-grant operational costs will be reduced [2206.03212]. Decolonial AI extends sustainability beyond project mechanics to “renewal of affective and political communities,” arguing that lasting decolonial change requires flows of funding, mentorship, and leadership opportunities reserved for communities historically dispossessed by algorithmic coloniality [2007.04068].

The NLP and African data ethics literatures emphasize infrastructural durability. NLP points to low-compute methods, democratized hardware access, and localizable open tooling as necessary responses to the “low-resource double bind” [2311.08391]. African data ethics formalizes the pacing problem through
$$
\frac{dI}{dt} \le \alpha S(t), \quad \alpha \in (0,1],
$$
where technical rollout should not outstrip social and institutional readiness [2502.16043].

Across these formulations, DMS is consistently presented not as a checklist completed at deployment, but as an ongoing ethical and organizational commitment. In computing education this commitment culminates in Indigenous-led resurgence; in AI it culminates in reciprocal, reflexive, solidarity-based practice; in NLP it requires intervention in data, algorithms, and software; and in data science it requires self-determination, communalism, and institution-building. The shared premise is that decolonial transformation is sustained only when epistemic change, material redesign, and governance redistribution proceed together.

Source: https://www.emergentmind.com/topics/decolonial-mindset-stack-dms