The New AI Colonialism: How Imported Algorithms Are Rewriting Power in the Global South

Artificial intelligence is touted as an advanced technology that can help governments improve public services, strengthen agriculture, expand financial inclusion, modernise healthcare, and make education systems smarter. Businesses are promised productivity gains while development agencies see AI as a tool for accelerating innovation in regions that have historically been excluded from technological transformation.

However, the spread of AI into postcolonial societies raises a deeper question: when powerful technologies travel across unequal global systems, do they reduce inequality or reproduce it?

A new review published in AI & Society argues that AI ethics cannot be understood only through familiar principles such as transparency, fairness, privacy, and accountability. In postcolonial contexts, the ethical risks of AI are tied to older structures of power: dependency on foreign technology, extraction of data and labour, imported governance standards, and the marginalisation of local knowledge.

The study, authored by Itoro Abraham of Lancaster University, synthesises 50 peer-reviewed articles published between 2019 and 2025. Drawing on postcolonial theory, it examines how AI systems shape ethical outcomes in societies marked by histories of colonial rule, economic dependency, and epistemic exclusion. Its central claim is blunt: AI ethics in these contexts is not merely a technical problem. It is a political-economic problem.

The review identifies four interlinked dynamics, algorithmic colonialism, data colonialism, platform imperialism, and platform sub-imperialism, that describe how AI systems can extend global and regional power asymmetries through infrastructure, data flows, labour arrangements, and governance frameworks.

Imported AI can carry hidden systems of dependency

The paper states that AI systems do not arrive as neutral tools; they arrive with infrastructures, standards, assumptions, and business models attached. In many postcolonial contexts, AI adoption depends on foreign-owned cloud services, proprietary software, externally designed models, and global platform ecosystems. Governments may draft national AI strategies, but much of the actual infrastructure, servers, data pipelines, model architectures, software platforms, and technical standards, remains controlled by multinational corporations or external actors.

This creates what the author describes as a condition of constrained or "liminal" sovereignty. States may formally regulate AI, but they often lack meaningful control over the systems that store data, process information, and shape algorithmic decision-making. The result is sovereignty on paper, dependency in practice.

The paper links this to "algorithmic colonialism," a process through which models, datasets, and design assumptions developed largely in Euro-American contexts are applied to postcolonial societies with limited sensitivity to local realities. An algorithm trained on data from one social world may misread another. It may classify communities inaccurately, ignore local languages, erase informal practices, or treat culturally specific realities as statistical anomalies.

These risks are especially serious when AI is deployed in high-stakes sectors such as healthcare, education, finance, policing, agriculture, and welfare administration. In such settings, algorithmic errors are not abstract. They can affect access to loans, diagnosis, public benefits, security scrutiny, or market opportunities.

The study also critiques "governance transfer," where AI ethics frameworks developed in the Global North are adopted globally with limited adaptation. Principles such as fairness and accountability may sound universal, but their meaning depends on institutional capacity, historical context, legal traditions, social inequality, and who gets to define ethical priorities.

The key point isn't to reject AI, but to ensure that countries adopt it thoughtfully, with clear attention to ownership, control, accountability, and local relevance. Without those questions, imported AI can become a new layer of dependency rather than a tool for autonomy.

Data, labour, and knowledge are becoming the new sites of extraction

The study further analyses AI as an extractive system. In colonial economies, land, minerals, labour, and raw materials were extracted from dominated territories for external accumulation. In the AI economy, the review argues, data, digital labour, and local knowledge can play a similar role.

Data colonialism occurs when information generated by people, communities, institutions, and markets in postcolonial settings is collected, processed, monetised, or governed elsewhere. Communities become sources of raw data, while the value produced from that data accumulates in external platforms, analytics firms, and cloud infrastructures. This raises urgent questions about data sovereignty. Who owns data generated in postcolonial societies? Who decides how it is used? Who benefits from AI systems trained on it? And what rights do communities have when their realities are translated into machine-readable assets?

The paper argues that ethical AI requires more than individual consent forms or privacy notices. It requires collective governance over data, especially when data reflects communities, languages, public institutions, health systems, or social behaviour.

The study also highlights labour opacity. AI is often marketed as automated intelligence, but many AI systems depend on vast amounts of human labour: data annotation, image labelling, content moderation, transcription, cleaning datasets, and verifying outputs. Much of this work is outsourced to low-paid or precarious workers, often in developing regions.

These workers are essential to AI systems but remain largely invisible in public debates about innovation. Their labour supports products used by global firms, yet they may face weak protections, poor wages, psychological harm from content moderation, and little recognition. The study therefore pushes AI ethics beyond algorithmic bias and into the labour conditions behind AI production.

Another concern is epistemic injustice, the marginalisation of local knowledge systems. When AI models privilege Western categories, languages, and institutional assumptions, they can undermine Indigenous and community-based ways of knowing. The ethical question is not only whether AI systems are accurate, but whose knowledge counts in their design.

Resistance is emerging, but it remains structurally constrained

The review identifies multiple forms of resistance, adaptation, and local agency. Some governments and institutions are seeking to localise AI strategies, build regional data infrastructure, and incorporate Indigenous or communitarian ethical frameworks. The study points to Ubuntu-informed approaches as one example of ethical thinking grounded in relationality, collective responsibility, dignity, and social repair rather than only individual rights.

There are also efforts to create sovereign data centres, regional cloud infrastructures, open-source tools, local datasets, and participatory AI governance mechanisms. These initiatives aim to reduce dependence on foreign platforms and make AI systems more responsive to local languages, institutions, and social priorities.

However, the study warns that resistance remains fragile. Many localisation efforts operate within global systems still dominated by external capital, standards, and platforms. Governments may adopt the language of data sovereignty while continuing to rely on foreign cloud providers. National AI strategies may invoke local values while using imported templates. Local firms may gain visibility while remaining dependent on global infrastructure.

The paper describes this tension through the idea of technological mimicry. Postcolonial actors may adopt external technologies and governance models to signal modernity, attract investment, or participate in global AI ecosystems. This can create opportunities for adaptation, but it can also reinforce dependency when imported models crowd out local innovation.

The study also introduces the concept of platform sub-imperialism, pointing to power inequalities within the Global South itself. Not all domination flows from North to South. Regional technology hubs, intermediary firms, and emerging platforms can also reproduce unequal relationships across neighbouring or less powerful contexts.

This vital corrective prevents the debate from becoming a simplistic story of "the West versus the rest." AI power is layered, multi-scalar, and increasingly regional. Ethical governance must therefore address both global asymmetries and intra-South inequalities.

Decolonising AI requires infrastructure, not just principles

AI ethics in postcolonial contexts cannot be fixed by adding more principles to existing frameworks. Principles matter, but they are insufficient if the infrastructure, labour relations, data flows, and governance systems remain unequal.

Governments must prioritise digital sovereignty with practical substance. This includes investment in regional data centres, interoperable systems, open-source tools, public-interest AI infrastructure, and procurement rules that prevent long-term lock-in to foreign platforms. Contracts with technology vendors should include data portability, transparency obligations, exit clauses, and local capacity-building requirements.

AI governance must become participatory. Ethics councils, algorithmic impact assessments, and regulatory bodies should include local communities, civil society, workers, domain experts, Indigenous knowledge holders, and affected groups. Without participation, AI ethics risks becoming a policy language spoken by elites while communities bear the consequences.

Data sovereignty must move from rhetoric to enforceable governance. Community data trusts, cooperative data models, benefit-sharing rules, and regional data protection frameworks can help rebalance power between local populations and global technology firms.

Labour must become central to AI ethics. Governments and international organisations should require transparency across AI supply chains, protect content moderators and data workers, regulate subcontracting, and recognise the human labour behind supposedly automated systems.

The paper warns development agencies and multilateral institutions against technology-first interventions. AI projects should not be judged only by efficiency, scale, or innovation. They should be assessed by whether they strengthen local autonomy, reduce dependency, protect workers, respect community knowledge, and build accountable institutions.

For businesses, responsible AI in postcolonial contexts requires more than compliance with global ethics guidelines. Firms must respect local data rights, disclose labour practices, avoid extractive platform arrangements, and co-design systems with affected communities.

The author cautions that the study is based on English-language scholarship, which means it may underrepresent Francophone, Lusophone, Arabic, Indigenous, and locally produced research. Its evidence base is also uneven across postcolonial regions, reflecting broader inequalities in research funding and publication visibility. Since it is a critical literature synthesis, it does not provide new field data from communities, developers, or workers directly affected by AI systems.