Three Places from Which Artificial Intelligence Is Being Contested

Three cases show how artificial intelligence redistributes labour, knowledge and power across South Asia, Africa and Latin America.

Gender in South Asia, cultural sovereignty in Africa and regional governance in Latin America

Artificial intelligence is often presented through figures for investment, computing capacity or new generations of models. Yet three recent developments show that its configuration also depends on much older social relations. What matters is who gains access to knowledge, which languages and values are considered translatable, and which institutions acquire the power to decide how these systems are used.

A diagnosis of gender and artificial intelligence in South Asia, a forum on culture and technology held in Ethiopia, and a regional roadmap for Latin America and the Caribbean allow us to observe three different scales of this process. The first concerns bodies and occupational trajectories. The second concerns the forms of knowledge and heritage that feed technological systems, or remain outside them. The third lies within the institutions attempting to regulate those systems.

These cases do not represent a homogeneous bloc called the Global South. Their histories, inequalities and institutional capacities differ. Read together, however, they shift attention away from the technological centres that dominate international discussion and towards the places where the social conditions of automation are being negotiated.

South Asia — Inequality Accumulates Before It Reaches the Algorithm

On 9 July 2026, UNESCO’s Regional Office in New Delhi presented the first diagnosis devoted specifically to the position of women within the artificial intelligence ecosystem of Bangladesh, Bhutan, India, Maldives, Nepal and Sri Lanka. The study was produced by the South Asia chapter of UNESCO Women for Ethical AI, created in 2025 and co-chaired by Amrita University and UNESCO.

The report follows trajectories from higher education into research, employment, entrepreneurship and technological leadership. Its main conclusion is that access to university does not automatically lead to equivalent participation in the areas where AI is developed.

Women account for roughly one third of STEM students across the region, but their presence declines in computer science, engineering and model development. Differences between countries are also considerable. In Maldives, women make up close to 78 per cent of university students, yet only 2 per cent of enrolments in science and technology. In Nepal, they represent 24 per cent of enrolments in computing and information technology. In Bangladesh, only one in five STEM graduates is a woman. India is approaching parity in the sciences, while engineering and technology remain clearly male-dominated.

Inequality is reproduced in research. A bibliometric analysis of AI publications from 2015 to 2025 shows that 71.76 per cent of articles from the six countries include at least one woman author. Women nevertheless occupy only 26 per cent of first-author or corresponding-author positions, which are commonly associated with scientific leadership. Men produce 2.7 times as many corresponding-author publications in India, eight times as many in Bangladesh and almost fourteen times as many in Maldives. Sri Lanka shows the smallest gap.

The disparity appears again within the technological value chain. Women are concentrated in data preparation, content moderation and digital services, while remaining under-represented in model development, where technical authority, intellectual property and the best-paid jobs accumulate. According to the LinkedIn data used in the study, women account for 29.92 per cent of users in India who report AI engineering skills, 20.57 per cent in Nepal and 15.64 per cent in Bangladesh.

Entrepreneurship follows the same pattern. In India, around 10 per cent of AI start-ups include at least one woman founder. Fewer than 1 per cent of the country’s approximately 10,000 angel investors are women. These figures do not simply describe unequal access. They show who has the resources to select projects, set priorities and transform a technical idea into economic infrastructure.

From an anthropological perspective, the so-called gender gap cannot be reduced to an individual lack of training. These trajectories are shaped by class, caste, religion, age, disability, rural or urban residence and the distribution of care work. Technology enters these relations and may extend them.

The study itself also has limitations. It brings together countries with very different economic structures, education systems and levels of digitalisation. It combines national statistics, bibliometric analysis, data from a professional platform and qualitative interviews whose representativeness is necessarily partial. Even so, it offers an unusually rich basis for observing how inequality moves from education into employment, and from employment into decisions about which systems are designed.

AI thus appears as a social division of labour. Some people produce and organise the data; others develop the models; others receive their decisions. These positions distribute more than wages. They also distribute visibility, authority and the ability to decide which problems deserve to be automated.

Africa — Preserving a Culture Can Also Mean Extracting It

From 22 to 26 June 2026, the AfriCreative Future Forum was held in Addis Ababa, focusing on peace, social cohesion and unity in the age of AI. The meeting brought together young people, cultural institutions, public bodies and participants from Africa, Latin America and Europe to discuss the relationships between artificial intelligence, creative industries, heritage, education and youth participation.

The forum also presented findings from Ethiopia’s AI Readiness Assessment. UNESCO’s methodology is intended to identify institutional capacity, public priorities and governance frameworks before technological adoption is expanded.

One of the forum’s central arguments was that African countries should not be confined to consuming technologies developed elsewhere. AI systems should also be built from African languages, cultures, knowledge systems and values. This orientation linked digital literacy and innovation to disinformation, heritage restitution and the right of African societies to tell their own histories.

The proposal opens a particularly fertile field for anthropology. Technology can assist with the digitisation of archives, the translation of languages, the documentation of cultural expressions or public access to collections. It can also detach those expressions from their contexts, turn community knowledge into training data and generate products without consent or benefit-sharing.

The tension is especially visible in intangible cultural heritage. A song, ritual narrative, craft technique or visual repertoire is not merely a unit of information. It may be governed by rules of transmission, belonging, age, kinship, gender or spiritual authority. When such material is digitised without recognising those relationships, situated heritage can be turned into a resource available to platforms, research institutions or companies.

The Addis Ababa forum did not produce a binding policy. Its importance lies in the agenda it formulates. It connects technological governance to questions that are often absent from technical documents. These include who has the right to tell a story, who may authorise its reuse and who receives the economic or symbolic value it generates.

ResearchGate’s academic radar provides a complementary case centred on Burkina Faso. A chapter by Harriet Deacon, Freda Owusu-Sekyere, Avril Joffe, Léonce Ki and other authors, published by Oxford University Press in June 2026, examines how public frameworks can help communities participate equitably in the benefits derived from the use of their cultural heritage by AI systems. Its contribution shifts the discussion from abstract preservation towards rights, licensing and the distribution of value.

Another study identified through ResearchGate, published by Jake Okechukwu Effoduh in Science and Public Policy, analyses the risk that African governance frameworks may reproduce external regulatory models with little adaptation to the continent’s histories, legal traditions and knowledge systems. Against this normative mimicry, it proposes epistemic sovereignty grounded in participatory legitimacy, knowledge pluralism and relational conceptions of personhood associated with Ubuntu.

In this context, digital sovereignty does not consist only of owning servers, data centres or domestic models. It also includes the capacity to establish limits, licences, collective rights and conditions of reciprocity around the knowledge incorporated into those systems. It further requires recognition that communities are not merely content providers, but subjects with the authority to decide how their heritage circulates and is transformed.

Latin America and the Caribbean — From Declarations to Mechanisms

On 25 and 26 June 2026, representatives from more than twenty countries took part in the Third Ministerial and High-Level Summit on the Ethics of Artificial Intelligence in Latin America and the Caribbean, held in Santo Domingo. The meeting was organised by UNESCO, the Government of the Dominican Republic, OGTIC, the Ministry of Foreign Affairs and the Development Bank of Latin America and the Caribbean, with support from the European Union.

The summit concluded with the Santo Domingo Declaration and a regional roadmap for 2026–2027. The document seeks to move from general principles towards mechanisms that can be implemented during the next period of work. It organises its priorities around governance and regulation, talent and the future of work, protection of vulnerable groups, environmental sustainability and technological infrastructure.

Its proposed outputs include a regional group of specialists on disinformation and AI, a response manual for disinformation incidents, regulatory design workshops, training for public policy-makers, gender-sensitive measures, a study on the effects of AI on children, a standard for measuring environmental impact and analyses of regional high-performance computing infrastructure.

The move towards implementation is significant. The region is no longer limited to listing ethical principles, but is beginning to build courses, standards, technical groups and co-ordination mechanisms. The summit also consolidated permanent structures for co-operation and proposed annual ministerial meetings.

This institutionalisation also makes a critical reading possible. The roadmap advocates open, participatory and multi-stakeholder processes, but leadership and monitoring rest mainly with national governments and intergovernmental organisations. Social participation appears as a general principle, while indicators, timetables and many operational decisions remain located within state structures.

The roadmap includes ethnicity and diversity within the protection of historically marginalised groups. UNESCO’s public communications explicitly mention Indigenous and Afro-descendant communities among those exposed to algorithmic risks. Yet these actors do not appear as specific rights-holders, or as the subjects of dedicated mechanisms, in the main operational outputs. Their presence is absorbed into broader categories of vulnerability.

The distinction matters. Latin America and the Caribbean contain important experiences of Indigenous data sovereignty, community technologies, language defence and resistance to automated racial discrimination. Treating their protagonists only as populations in need of protection can obscure their ability to produce their own norms, knowledge and infrastructures.

Specialist coverage of the summit has also pointed to the absence of a more explicit formulation of regional technological sovereignty. It has further warned that the uncritical adoption of generative systems in schools, media and public administration may homogenise thought, weaken historical memory and amplify disinformation.

ResearchGate provides a further academic counterpoint. Miguel Ángel Juárez Merino’s work on relations between the state and citizens in Latin American algorithmic governance identifies a persistent tension between the adoption of advanced ethical frameworks and the difficulty of translating them into practical guarantees. His analysis is useful for observing how AI can reinforce asymmetries already present in public administration, particularly when it intervenes in health, education, security, benefits or access to rights.

Anthropology can contribute by following these policies once they leave the declaration and reach a school, social registry, border, police station or health system. In those settings, terms such as inclusion, transparency, efficiency or risk take on concrete forms and produce different consequences according to the social position of each person or community.

Three Scales of the Same Transformation

The three cases reveal a sequence.

In South Asia, the issue concerns who can move along the chain from education to technological leadership. In Africa, it concerns the knowledge incorporated into systems and the conditions under which it may be reused. In Latin America and the Caribbean, it concerns the institutional devices intended to turn ethics into rules, public capacity, standards and monitoring mechanisms.

These scales are connected. Inclusive governance is difficult to build when the people developing technology come from highly restricted social groups. Cultural representation is not resolved by adding more data if communities lose control over it. An ethical declaration has limited reach when it lacks procedures for contesting decisions, demanding explanations or repairing harm.

Artificial intelligence does not arrive in empty societies. It enters divisions of labour, property regimes, colonial histories, linguistic hierarchies and different understandings of personhood and community. These three recent developments show that its future is being negotiated far from the technological centres that usually dominate international coverage.

What is at stake is not only who will gain access to artificial intelligence. It is also who will be able to define it, which forms of knowledge it may use, under what conditions and for which collective purposes.

References

UNESCO, CAF and OGTIC. (2026, 25–26 June). Santo Domingo Declaration.

Hirschfeld, D. (2026, 6 July). Grupo regional analizará desinformación e Inteligencia Artificial. SciDev.Net.

OGTIC. (2026). República Dominicana asume la Presidencia del Grupo de Trabajo Regional sobre Inteligencia Artificial.

Juárez Merino, M. Á. (2026). AI and Governance: State-Citizen Dynamics in Latin America. The International Journal of Technology, Knowledge, and Society. https://doi.org/10.18848/1832-3669/CGP/A1024

Note on Sources

ResearchGate is used here as a complementary academic radar. Wherever possible, dates, data and principal references have been checked against publishers’ pages, DOI records, institutional repositories and official sources. Its inclusion does not turn an academic publication into a news item of the day; rather, it helps place recent events within broader research debates.