Source-linked AI summary
Decolonial AI: Decolonial Theory as Sociotechnical Foresight in Artificial Intelligence
Shakir Mohamed, Marie-Therese Png, William Isaac
TL;DR
AI advances can reproduce harms through embedded values and asymmetrical power, especially for vulnerable peoples. The paper uses decolonial theory and cases of algorithmic coloniality to propose critical foresight and practices for more ethically aligned AI. It concludes that these perspectives can centre vulnerable communities and strengthen AI’s social contract.
Problem
AI ethics and power analysis can overlook historical continuities, local knowledge, and the disproportionate harms advanced systems impose on vulnerable peoples.
Method
The paper applies decolonial theory as ethical foresight, examines cases of coloniality, and proposes critical technical practice, reverse tutelage, and renewed affective and political communities.
Results
The paper identifies algorithmic oppression, exploitation, and dispossession across AI applications and argues that decolonial perspectives reveal power relations and colonial continuities in AI.
Takeaways & Limitations
AI communities should develop foresight, inclusive dialogue, and technical practices that give marginalised groups meaningful influence over AI development.
Takeaways & Limitations
The paper cautions that metropole-periphery binaries, grand historical narratives, and speaking for the oppressed can oversimplify decolonial analysis.
Abstract
from arXiv · showhide
This paper explores the important role of critical science, and in particular of post-colonial and decolonial theories, in understanding and shaping the ongoing advances in artificial intelligence. Artificial Intelligence (AI) is viewed as amongst the technological advances that will reshape modern societies and their relations. Whilst the design and deployment of systems that continually adapt holds the promise of far-reaching positive change, they simultaneously pose significant risks, especially to already vulnerable peoples. Values and power are central to this discussion. Decolonial theories use historical hindsight to explain patterns of power that shape our intellectual, political, economic, and social world. By embedding a decolonial critical approach within its technical practice, AI communities can develop foresight and tactics that can better align research and technology development with established ethical principles, centring vulnerable peoples who continue to bear the brunt of negative impacts of innovation and scientific progress. We highlight problematic applications that are instances of coloniality, and using a decolonial lens, submit three tactics that can form a decolonial field of artificial intelligence: creating a critical technical practice of AI, seeking reverse tutelage and reverse pedagogies, and the renewal of affective and political communities. The years ahead will usher in a wave of new scientific breakthroughs and technologies driven by AI research, making it incumbent upon AI communities to strengthen the social contract through ethical foresight and the multiplicity of intellectual perspectives available to us; ultimately supporting future technologies that enable greater well-being, with the goal of beneficence and justice for all.
1 How Values Shape Scientific Knowledge and Technology
AI increasingly shapes cultural, economic, and political life, while its values and power relations raise questions about harms, accountability, and ethical foresight.
- AI should be understood both as technological artefacts and as networks and institutions shaping society.
- Advances driven by computation, data, and learning algorithms have produced benefits alongside opportunities for misuse beyond designers’ expectations.
- Values in science and technology include both epistemic standards and contextual moral, societal, and personal concerns.
- AI ethics frameworks emerged to reorient power relations after unethical research, but early AI guidelines underrepresented harms first felt by developing-country conflict zones.
- A healthcare prediction algorithm used healthcare expenditure as a proxy for need, rejecting African-American patients at higher rates and exacerbating structural inequities.
- AI’s ability to rapidly ingest, perpetuate, and legitimize bias warrants reappraisal of its ethical and socially beneficial use.
- Critical science examines cultural assumptions, values, and power among stakeholders and technological artefacts, while decolonial perspectives illuminate exploitation, bias, and dispossession.
- The paper presents decolonial theory as ethical foresight for surfacing AI blind spots, limitations, and power relations while empowering vulnerable peoples.
2 Coloniality and Decolonial Theory
Decolonial theory distinguishes enduring coloniality from historical colonialism and offers perspectives for interrogating dominant knowledge, values, and power in contemporary AI.
- Coloniality describes power dynamics that survive colonialism through historical dispossession, enslavement, appropriation, and extraction.
- Structural decolonisation seeks to undo contemporary mechanisms of power, economics, language, culture, and thinking by questioning dominant knowledge and values.
- Decentring decolonisation rejects imitation of the West and re-centres knowledge on distinct identities, histories, problems, and solutions.
- An additive-inclusive view supports alternative approaches, localisation, pluriversalism, and environments where new ways of creating knowledge can flourish.
- An engagement view examines science from the margins, asking who is included, excluded, silenced, and served by its applications.
- Decolonial analysis maps centres of power and less-powerful peripheries while connecting colonial histories to present-day underdevelopment.
- The paper cautions that metropole-periphery binaries can oversimplify lived experience and that grand narratives and speaking for the oppressed are theoretical pitfalls.
3 Algorithmic Coloniality
The paper applies decolonial theory to AI by identifying digital and algorithmic continuities of colonial power and organizing them into a foresight taxonomy.
- Digital territories can become sites of extraction and exploitation, while persistent unquestioned institutions and values reproduce structural coloniality in AI.
- Data colonialism and data capitalism frame data as a material resource through which historical continuity and economic expansion operate.
- Algorithmic coloniality concerns algorithms’ effects on resource allocation, sociocultural and political behaviour, discriminatory systems, labour markets, geopolitics, and ethics discourse.
- The paper introduces a decolonial foresight taxonomy comprising institutionalised algorithmic oppression, algorithmic exploitation, and algorithmic dispossession.
- Its sites of coloniality include algorithmic decision systems, ghost work, beta-testing, national policies, and international social development, showing why AI power analysis cannot be ahistorical.
Site 1: Algorithmic Decision Systems
Algorithmic decision systems can reproduce historical and structural inequities across criminal justice, surveillance, identity, employment, healthcare, and everyday interactions. A decolonial framework situates these harms within wider systems of racial capitalism, class inequality, and heteronormative patriarchy, while supporting geographically expansive oversight and redress.
- Site 1: Algorithmic Decision Systems: Predictive systems presented as evidence-driven and unbiased can instead entrench historical injustice and amplify existing inequities.The paper identifies these systems across policing, surveillance, government services, identity, and speech.
- Site 1: Algorithmic Decision Systems: Documented harms include discriminatory policing, facial-recognition failures, toxic-language classifications, recruitment discrimination, trans-targeting, and facial inference of queerness.These examples span criminal justice and everyday interactions.
- Site 1: Algorithmic Decision Systems: Algorithmic inequities should be historically contextualised within racial capitalism, class inequality, and heteronormative patriarchy rooted in colonial history.The paper links automated institutional harms to histories of racist expropriation.
- Site 1: Algorithmic Decision Systems: A decolonial framework connects algorithmic oppression across geographies and supports analysis beyond North American or European definitions of identity and harm.The paper associates this broader analysis with inclusive oversight and redress mechanisms designed from the start.
Site 2: Ghost Workers
AI development depends on globally distributed human annotation, including work performed by economically vulnerable people and prisoners. A decolonial lens interprets this labour through colonial continuities in extraction, regulation, and cross-border outsourcing.
- Site 2: Ghost Workers: Large volumes of AI data require human annotation, often performed by globally distributed workers, prisoners, and economically vulnerable people.The paper calls these annotators “ghost workers.”
- Site 2: Ghost Workers: Distributed annotation can enable economic development and flexibility while also involving very low pay and limited attention to working conditions, support, and safety.The passage presents these as competing aspects of the same labour arrangement.
- Site 2: Ghost Workers: Cross-border outsourcing allows AI industries to reorganize production around post-colonial economic inequalities, reducing costs and enhancing corporate profit.The paper connects ghost-worker locations in previously colonised geographies with continuing extraction and exploitation.
Site 3: Beta-testing
Beta-testing can turn marginalised populations and weakly regulated countries into testing grounds for predictive systems. A decolonial historical lens frames these practices as continuities of scientific exploitation and raises questions about accountability, responsibility, contestation, and recourse.
- Site 3: Beta-testing: Scientific and technological progress has repeatedly relied on experimentation involving marginalised populations, including colonial subjects and African Americans.The paper uses this history as the context for examining contemporary software beta-testing.
- Site 3: Beta-testing: Beta-testing is the testing and fine-tuning of early software versions with real users and use-cases to identify usage issues.The paper examines predictive-system testing through this practice.
- Site 3: Beta-testing: Ethics dumping exports harms and unethical research practices to vulnerable populations or low- and middle-income countries, often along colonial fault lines.Ethics shirking describes protections omitted when harms exceed what firms are required to do.
- Site 3: Beta-testing: Cambridge Analytica beta-tested election tools in Kenya and Nigeria partly because their data-protection laws were weaker than those in the United Kingdom.The systems later interfered in electoral processes and worked against social cohesion.
- Site 3: Beta-testing: Predictive child-welfare systems initially targeted Māori communities, while Palantir experimentally deployed police-surveillance algorithms in New Orleans without public approval.The New Orleans deployment disproportionately impacted African-Americans, and the paper places these cases within longer histories of institutional racism.
- Site 3: Beta-testing: Decolonial analysis raises questions of accountability, responsibility, contestation, and recourse where low regulation combines with limited local expertise and historical knowledge.The paper also warns against neglecting ethical variation across identity and geography.
Site 4: National Policies and AI Governance
Global AI governance is shaped by uneven representation and power, with economically developed countries exerting disproportionate influence over ethics guidelines, policies, and standards. Decolonial theory offers a framework for examining dependencies, infrastructure ownership, and the distribution of risks and benefits.
- Site 4: National Policies and AI Governance: Global AI governance debates raise questions about who regulatory norms protect, who projects them, and who benefits from centralised power and capital.The paper identifies data inequality and infrastructure sovereignty as part of these power imbalances.
- Site 4: National Policies and AI Governance: Africa, South and Central America, and Central Asia are under-represented in global AI ethics debates, while economically developed countries shape the discussion more strongly.The paper associates this imbalance with neglect of local knowledge, cultural pluralism, and global fairness.
- Site 4: National Policies and AI Governance: India, Indonesia, and South Africa refused to sign the 2019 Osaka Track declaration because their interests, concerns, and priorities were not represented.The paper identifies similar concerns regarding the OECD AI Principles.
- Site 4: National Policies and AI Governance: A metropole-periphery model highlights how governments and industries can impose normative values and standards while forestalling alternative visions.It also draws attention to the representation of resource-constrained countries in governance processes.
- Site 4: National Policies and AI Governance: Decolonial theory helps policymakers interrogate power imbalances, structural dependencies, critical-data-infrastructure ownership, and unequal distributions of AI risks and economic benefits.The framework applies across policy discourse and computational-technology design, development, and deployment.
Site 5: International Social Development
AI-for-development discourse presents advanced technologies as solutions to complex scenarios, while decolonial critiques highlight dependency, dispossession, and ethics dumping. Co-development and reflexive evaluation are proposed to improve AI’s sociopolitical, economic, linguistic, and cultural relevance while shifting power asymmetries.
- AI-for-development discourse often presents advanced technologies as solutions for complex developmental scenarios.
- Decolonial critiques identify dependency, dispossession, and ethics dumping as recurring concerns in technology-for-development projects.
- Co-development can support AI systems’ sociopolitical, economic, linguistic, and cultural relevance across communities while shifting power asymmetries.
- A decolonial view promotes reflexive evaluation of cultural encounters and sustained questioning of development’s philosophical basis.
4 Tactics for a Decolonial AI
The paper proposes three tactics for decolonial AI: critical technical practice, reciprocal engagements and reverse pedagogies, and renewed affective and political communities. These tactics address coloniality in AI’s applications and supporting structures by questioning assumptions, enabling reciprocal knowledge practices, and strengthening affected communities’ influence.
- Decolonial AI treats applications, data, networks, and policies as expressions of coloniality requiring the decolonisation of power.
- The paper submits three tactics: critical technical practice, reciprocal engagements and reverse pedagogies, and renewed affective and political community.
- Critical Technical Practice: Critical technical practice combines technical development with reflexive criticism of hidden assumptions and politically situated values.
- Critical Technical Practice: The practice applies critical scrutiny to fairness, safety, diversity, policy, and resistance rather than treating ethical assessment as a secondary task.
- Reciprocal Engagements and Reverse Tutelage: Reciprocal tutelage can be enacted through dialogue, documentation, and design, allowing knowledge from colonised peoples and peripheral communities to shape AI.
- Reciprocal Engagements and Reverse Tutelage: Reverse tutelage challenges the view that knowledge and data provide complete abstractions, emphasizing selection and interpretation under differing value systems.
- Renewal of Affective and Political Community: Political and affective communities are central because AI governance depends on inclusion, ownership, contestation, redress, and the ability to reverse technological interventions.
5 Conclusions
The paper presents decolonial thinking as a critical resource for questioning inherited hierarchies, power relations, and cultural assumptions in AI. It calls for operational methods, inclusive stakeholder dialogue, and ethical foresight to support responsible AI oriented toward beneficence and justice.
- Decolonial thinking expands practitioners’ ability to question AI’s developers, locations, cultural assumptions, and embedded power relations.
- Operationalising critical AI practice requires foresight, case studies, new research cultures, and technical research in fairness, value alignment, privacy, and interpretability.
- Inclusive dialogue should give marginalised groups meaningful avenues to influence AI decision-making and avoid predatory inclusion, oppression, exploitation, and dispossession.
- The paper positions ethical foresight and multiple intellectual perspectives as resources for strengthening AI communities’ social contract toward beneficence and justice.