Artificial intelligence for development (AI4D): A contested notion

Abstract

Recently, the notion of artificial intelligence for development (AI4D) has been mobilized by various actors in the global South and North. We identify five analytical categories to help us understand the different and often contested perspectives on AI4D. They are (a) a developmentalist framework that emphasizes discourses around modernity and progress through a technoliberal lens of ‘catching up’; (b) an economic development framework taken up by African states, private sector and civil society, highlighting a positive and more future-looking outlook on AI's potential for development; (c) an international policy framework tied to globally agreed on policies such as the Sustainable Development Goals; (d) a colonial and extractivist framework that articulates how AI4D reinforces old processes of oppression in new ways; and (e) decolonial AI discourses grounded in Latin American, African and Indigenous approaches. Our critical review of literature on AI4D and related expressions shows that while the notion applies broadly to the global South, the majority of publications use the term in reference to AI development on the African continent. This commentary enriches our understanding of the plurality of meanings, where they come from, what they do, and what they leave unaddressed.

This article is a part of special theme on Datafied Development. To see a full list of all articles in this special theme, please click here: https://journals.sagepub.com/page/bds/datafied_development?pbEditor=true

Introduction

The notion of artificial intelligence for development (AI4D) gained increased adoption at the onset of the COVID-19 pandemic. This coincided with the 2020 launch of AI4D Africa, a program funded by Canada and Sweden, which aims to support an applied AI research agenda to address development challenges in Africa (IDRC, n.d.). Since then, the term has been mobilized by various actors, including academia (Mann and Hilbert, 2020), civil society (Borokini et al., 2022), corporations (Microsoft, 2024), international institutions (Pizzi et al., 2020) and governments (IDRC, n.d.). In this research commentary, we shed light on the different meanings behind this notion, where they come from, what they do and what they leave unaddressed.

We have undertaken a broad and critical review of some of the writings on AI4D. We found over 55 publications including articles, chapters, books and reports using the term AI4D and related expressions. While AI4D applies broadly to the global South, we found that the majority of publications reviewed use the term in reference to AI development on the African continent. From this material, we have identified five analytical categories to understand the different and often contested perspectives on AI4D along with their underlying assumptions. They are: (a) a developmentalist framework (16 publications), (b) an economic development framework (12 publications), (c) an international policy framework (11 publications), (d) a colonial and extractivist framework (9 publications), and (e) a decolonial framework (8 publications).

A developmentalist framework

The first category we explore is what we call a developmentalist framework about AI4D and a technoliberal lens of ‘catching up’ that is believed to alleviate ills such as poverty with AI. This work harks back to Truman's 1949 state of the union address signaling the beginning of a new communication for development era where the United States assumed an imperial role through new engagements in international development assistance and advancements in science and technology. New discourses on Information and Communication Technologies for Development (ICT4D) emerged in the late 1990s, promising to solve long-standing development issues by bridging the digital divide between the wired and the unwired (Mazzarella, 2010). Present debates on AI4D carry many of the same ICT4D promises and quasi-magical qualities stressing how technologies, if used responsibly, can alleviate social, economic, and political ills in the South and democratize access. However, AI4D extends beyond these goals due to the complex entanglements between infrastructures, data, algorithms, and people. As a new international development actor, AI is shaping the governance of people and countries through data extraction and labor.

Under a developmentalist framework, we note that the prevailing belief is that the African continent is “lagging behind most of the world” (Gwagwa et al., 2021: 3) in terms of AI advancement. The urgency of catching up with the West is further underscored by the Government AI Readiness Index (Oxford Insights, 2023), which categorizes Africa (excluding North Africa) as unprepared, given its lowest score. Here, we see the old trope of needing to catch up. From this perspective, countries can only stand to gain from assistance provided by international development actors and technology firms to further their AI readiness.

Critical of a “ready-to-wear” style development, Felwine Sarr (2016) considers this way of thinking as a Western endeavor to impose its myths and epistemologies and make itself the aspiration for other nations. Only by unpacking the developmentalist discourse around AI4D can we reveal the long history of Western-style development that has consistently denied the autonomy of nations in pursuing diverse trajectories and pathways to respond to their own challenges.

An economic development framework

The second category of discourse we identify presents AI4D as a mechanism to achieve social and economic development goals, with a strong emphasis on its application in Africa. It is tempting to view the developmentalist and economic development frameworks as two sides of the same coin given their shared roots in the 1950s modernist approach which saw technology as a vehicle for developing countries to achieve progress. However, discourses under the economic development framework are distinct in their positive and forward-looking outlook.

In our analyses, AI4D is presented through nationalist perspectives where AI can transform key sectors including agriculture, finance, education and healthcare (Arakpogun et al., 2021; Borokini et al., 2022) and create jobs. Many of the work in this framework present AI as inevitable and speaks of how African countries can embrace emerging technologies to “leapfrog” legacy systems to development and economic growth (Ndemo et al., 2019), suggesting that AI solutions in Africa could be a “game-changer” (Arakpogun et al., 2021).

The publications in this category mobilize the discourse around leapfrogging into the modern world and reaping the economic benefits of new data and AI economies thanks to narratives facilitated by the Fourth Industrial Revolution (4IR). This is well expressed in Cyril Ramaphosa's (2020) remarks on South Africa's AI national strategy where he discusses the “great tectonic shift” towards an economy based on the 4IR that prioritizes “inclusive growth” and aims to “propel new solutions to our developmental challenges” (para. 2). Yousif Hassan (2024) draws our attention to the diverse imaginaries informing AI practices across the African continent, which is indicative of a rupture with the past and present and a more future-looking approach to the role that AI can and should play in the lives of Africans.

In the literature, AI is described as an important opportunity and a potential driver of growth, particularly for governments. Through the case study of the GovChat, Plantinga et al. (2019), examine how such AI-enabled projects can improve public services in South Africa and help marginalized communities access local government. In Kenya, the government-commissioned Distributed Ledger and Artificial Intelligence Taskforce refers to emerging technologies as opportunities for the country to become “a leading developed economy by leveraging and taking ownership of the Fourth Industrial Revolution” (Ndemo et al., 2019: 08).

The legacy of colonialism receives limited attention under this framework, as delayed (or lack of) technological development is characterized as a “missed opportunity” rather than as remnants of a historical and ongoing exploitation of the continent's people and resources. Narratives under this framework overlook a crucial aspect of AI development: Who stands to gain from this economic push, and who stands to lose out?

An international policy framework

This framework offers a conceptualization of AI4D tied to globally agreed on policies, particularly the United Nations Sustainable Development Goals (SDGs). In recent years, UN agencies have shown more interest in using AI to advance their development work, including the launch of the AI for Good Global Summit. Here, AI is depicted as a vehicle to speed up the process of solving complex issues and attaining the development goals.

Publications under this framework often use a celebratory tone in their capacity to reach internationally agreed goals and society's transformation thanks to AI. As a case in point, Goralski and Tan (2020) suggest that AI will bring business opportunities, enable effective and efficient public policy for sustainability, and improve access, connectivity, and efficiency in the areas of healthcare, education, farming and transportation. Through their investigation of 24 case studies of the use of AI in Latin America and Africa, Mann and Hilbert (2020) affirm that these projects provide evidence for bright opportunities to foster the development agenda, provided that tensions between global efficiency and local needs are addressed.

What is frequently overlooked in such discussions is an exploration of the political economy underlying the SDGs themselves. The SDGs are based on the widest consensus possible among UN member countries thereby reinforcing what Telleria and Garcia-Arias (2022) call the status quo instead of bringing meaningful transformation. What remains hidden in the discourse on the integration of AI for achieving the SDGs is that, according to some, the development goals themselves reinforce a neoliberal development paradigm characterized by privatization and marketization (Telleria and Garcia-Arias, 2022). The promotion of the free market and the reduction of state involvement in the majority world are embedded in ideas about AI4D. These ideas focus on the “automation” of development issues which are, in turn, a response to the destruction of global South countries due to ongoing structural adjustment programs (SAPs). Gurumurthy and Chami (2019: 01) argue that “AI as an essential ingredient” in a global capitalist system “intensifies an already unequal and unfair international development context.” Rather than making visible the economic, environmental and social harm inherent in the assumptions behind mainstream understanding of SDGs, the international policy framework creates an illusion that “no one is left behind” not even the planet. To repoliticize the policy discourse on AI4D and its repercussions on the majority world, it is imperative to reveal these underlying assumptions.

A colonial and extractivist framework

The fourth framework situates AI4D as part of a colonial extractive project. It reveals how AI4D and its underlying assumptions perpetuate established forms of dominance in the majority world. This perspective unveils the extraction, and exploitation inherent in the global AI industry. It gives due consideration to the historical aspects of these practices and how new power dynamics that persist in AI reinforce older processes of domination (Birhane, 2020; Mano, 2023).

AI4D is perceived as a potential instrument for the recolonization of Africa using countries and their populations as testing grounds. Birhane (2020) introduces the concept of algorithmic colonization to elucidate how contemporary actions of Western technology companies mirror the exploitation observed in the colonial era. Under the guise of well-meaning AI-driven solutions, technological developments imbued with Western values lead to the displacement of local products as African countries become dependent on Western-imported digital infrastructures, leading to what some have described as the “new scramble for Africa” (Mano, 2023).

The surveyed scholarship draws clear connections between AI development and how global majority countries have been consistently depleted of their resources and human strength through sophisticated technological and data-driven means. Benyera (2021) questions the promises of the 4IR, arguing that Africans will benefit the least and it will only serve as another form of recolonization for Africa. He points to new forms of dependency created by five large tech monopiles (Facebook, Apple, Google, Amazon, and Microsoft) that make up what he calls the “emerging tech oligarchy” (Benyera, 2021).

This framework also brings into focus Africa, Asia and Latin America's position at the initial stages of the value chain. It highlights the extraction of raw materials needed to build AI technologies, the precarious work carried out by data annotators tasked with “cleaning” datasets and making our AI experience less toxic (Crawford, 2021) and how communities stand to lose as old forms of exploitation and domination are replicated for the development of new AI products, services, and industries (Ngwane and Tshoaedi, 2021). Arun (2020) highlights how India serves as both a source of data and users of tech services where the “commodification of citizens” leads to a prioritization of markets over the data rights of citizens.

Framing AI4D as a manifestation of digital colonialism has helped nourish the emergence of critical discourses and practices in the field. A decolonial reorientation has given rise to thinking about AI differently. This is the focus of our last category.

A decolonial framework

The fifth category that we examine takes the “D” in AI4D to signify decolonial, encapsulating what scholars refer to as decolonial AI. The growing interest in the decolonial turn within the field of AI spearheaded by feminist, indigenous and scholars of color often from the global South has created a space of creativity for thinking about AI from decolonial inflections.

The publications surveyed in this category are rooted in a plurality of approaches (Latin American, African, and Indigenous) in their understanding of decolonial thoughts. Mohamed et al. (2020) explore decolonial AI by drawing insights from Latin American scholars and redirecting the conversation from biases to the intersection of values, power, and AI. Rachel Adams (2021) contributes to this discourse by bringing together decolonial aspects of AI and African scholarship. Adams concludes her piece by asking if it is at all possible to decolonize AI and suggests that envisioning the future in decolonial terms also means imagining a world without AI (2021). Indigenous decolonial orientations have also led to novel ways of thinking and doing AI. Indigenous AI serves as a means to affirm indigenous sovereignty and calls for communities to assert control over AI. It is a strategy to resist the imposition of settler colonial developmentalist ideas that have long been imposed on indigenous peoples. Indigenous AI is concerned with challenging Western rationalist epistemologies which limit “imagination, frameworks, and language to effectively engage alone with the new ontologies created by future generations of computational systems” (Lewis et al., 2020: 6).

Discursively and materially, the narratives and practices on decolonial AI offer refreshing and much needed re-orientations to think about AI4D differently. Yet, a nuanced examination into the various interpretations of the term decolonial is essential. Is the term decolonial merely becoming interchangeable with development, or does it truly embody a transformative potential for reshaping our world?

Conclusion

In this commentary, we have demonstrated that the plurality of discourses surrounding AI4D are founded on assumptions that require thorough examination. In much of the material we reviewed, the term tends to obscure aspects of unrestrained capitalist development, the unquestioned embrace of techno-economic dominance to bring about socio-economic “progress,” the western hegemony on AI4D, the environmental degradation caused by AI, and the invisible exploitation of gig workers. It is only by understanding the historical, social, economic, and ideological foundations of each discourse that scholars and practitioners will be able to grasp what this term does and how it shapes AI in international development.

Finally, we highlighted the extent to which AI4D is a space for contestation. Prioritizing decolonial perspectives, the well-being of workers, communities and the planet can override the geopolitical agendas and profitmaking of industries and governments. These perspectives help us consider a world where communities decide how resources (financial, human, environmental) are used, including the right to refuse AI. The framework of refusal as a political orientation implies a desire for autonomy, self-determination, and sovereignty, all principles that are constitutive of a political tradition that the discussions on AI4D are in great need of.