Artificial intelligence and development projects: A case study in funding mechanisms to optimise research excellence in sub-Saharan Africa

J. Shawe-Taylor
University College London
United Kingdom
D. Orlič
Knowledge 4 All Foundation
United Kingdom

Artificial intelligence (AI) has been attracting increased attention from researchers, entrepreneurs, investors and policy makers on all continents. Innovative national and international development collaboration mechanisms, such as micro-funding and social impact bonds, are being tested, refined and implemented to assist AI researchers, contribute to scientific excellence and scale to market. This essay looks at emerging networks of excellence in the Global South, particularly AI4D Africa. It examines how bottom-approach, small-scale investments resulted in significant research on different scientific and non-scientific, engineering and educational topics, including a profile of African languages.

Since sustainable development is a challenge for all countries and in different ways, many are developing their own approaches to using AI to help address their specific needs. Some of these approaches span different continents and regions, and others are location-specific. For example, a research group in one region might use satellite imagery to understand the quality of water resources in lakes. A group in another region might analyse news items in a number of different languages monitoring the same issue. Yet both groups are addressing water quality management, one of the Sustainable Development Goals (SDGs) (UN, 2022).

An opportunity exists to connect research groups and tap into different solutions, case studies, skills and competences via co-operation mechanisms such as networks and centres of excellence in AI and sustainability (NAIXUS, 2022). Most of these mechanisms used to boost AI in science in developing countries are having a positive impact, while being flexible and cost effective.

The introduction of AI applications in the Global South makes possible innovative, data-driven, technical innovations to help address pressing socio-economic problems and improve policy actions. AI can facilitate scientific breakthroughs, improve medical diagnoses, increase agricultural productivity, optimise supply chains and help equalise development of skills through highly personalised learning. However, AI could also widen the breach between developed and developing countries’ capabilities in science.

Most AI experts work in North America, Europe and Asia, with sub-Saharan Africa barely represented in the global pool of experts. In many new AI initiatives in Africa, the expertise involved is barely visible in developed-country technology hubs. Nevertheless, a significant AI community has grown up in Africa in recent years, with initiatives such as Deep Learning Indaba2022, Masakhane Foundation2022, Data Science Africa (DSA, 2022) and Data Science Nigeria (DSN, 2022) with many more establishing legal entities to formalise their work. Such bottom-up approaches can bypass burdensome and bureaucratic top-down university co-operation systems.

These self-mobilising and unique emerging expert communities in Africa have created novel opportunities for co-operation and innovation by introducing funding for a range of micro-scale research projects. This funding model assumes that large projects can be cumbersome and present significant bureaucratic bottlenecks; a dynamic framework of targeted and quickly disbursed financial support is more effective overall. The question is whether this approach can accelerate innovation at scale.

Networks of excellence in science aim to strengthen particular areas of science and technology through collaboration. Some networks operate at European level and aim to marshal the resources and expertise needed for Europe to be a world force in a given field. PASCAL2 was the most ambitious of a number of European-funded networks of excellence in the fields of pattern analysis, statistical modelling and computational learning. It ran from 2008 until 2013 (CORDIS 2022).

The PASCAL2 organisational and financial model helped inspire more recent networks of European AI researchers. These include the European Learning and Intelligent Systems Excellence initiative (ELISE, 2022). and the European Network of Human-Centered Artificial Intelligence (Humane AI, 2022). It also includes networks outside Europe, such as AI for Development Africa, a network of researchers and practitioners in sub-Saharan Africa (AI4D, 2022).

These networks aim to encourage researchers to collaborate, to think outside of their particular research interests and help take machine learning into other fields. Early experiences with these networks offered important lessons. They had remarkable success, for example, in empowering and trusting people with the freedom to do research, with little or no funding up front and without constant pressure for results. PASCAL2 established the Knowledge 4 All Foundation (K4A) as a UK charity (K4A, 2022). Through this legacy organisation, it would make success stories, incentive models and methodologies explored in PASCAL permanently available in Europe and beyond.

K4A supports specialised scientific communities and the general public across the world with capacity building, open educational resources and education technologies. In collaboration with the Jožef Stefan Institute, for example, the Foundation has used VideoLectures (2022) to improve access to content in all subcategories of computer science. An award-winning open educational resource, VideoLectures has a tail of free machine-learning lectures dating to 2003. It is an example of a major enabler of AI uptake that can make AI tools and education accessible globally through the Internet.

In 2018, the Knowledge 4 All Foundation worked with UNESCO and Canada’s International Development Research Centre (IDRC) to map the landscape of AI in emerging economies. The mapping covered 33 countries and 617 institutions spread across Asia, Latin America and the Caribbean, the Middle East and North Africa, and sub-Saharan Africa. The result – the Emerging Economies Artificial Intelligence Ecosystem Directory (K4A, 2022) – was one of the first bottom-up mappings of AI entities in the Global South. It provided a basis for types of capacity building – from policies to support AI in science and funding bodies to research methods, dissemination of information on use cases, deployment, exploration, exploitation and operability. These could be applied to topics relevant to the SDGs. The results helped bootstrap a series of AI-related research and development initiatives in sub-Saharan Africa.

As a result of this work, IDRC provided funding to K4A in 2019 to help establish the AI4D network. This second project aimed to strengthen and develop a community of scientific and technological excellence in a range of AI-related areas. AI4D Africa developed a network of institutions and individuals working on and researching AI from across sub-Saharan Africa, via workshops and consultations. It delivered an AI-related research agenda (Gwagwa et al., 2021) with a focus on ethical, legal and social issues underpinning scientific quality in AI research. It also generated an AI capacity building agenda (Butcher et al., 2021) via a survey of universities. Further, it issued a call for multidisciplinary innovation projects within and outside the network, exploring local frontiers of research in AI.

K4A has helped co-ordinate initiatives that are already having a significant impact in the region. These include COVID-19 data challenges; a fellowship yielding 30 African language datasets covering 22 countries with 300 million speakers; a text-to-speech platform for African languages; and a registry of AI hot spots in Africa and engagement across many researchers and research institutions.

K4A designed two calls for applications for funding of micro-projects in 2019 and 2020. Projects were required to: 1) create a dataset; 2) have a novel and motivated goal; 3) involve a challenging yet manageable task with a scalable long-term vision; and 4) be accessible to the general public and researchers. The successful projects came from nine countries and encompassed AI applications in a diverse range of scientific and social objectives.1 Elected projects were awarded between USD 5 000-8 000 each. The call for micro-projects also generated the first African Grand Challenge in AI. It focused on curing leishmaniasis, a neglected disease that affects the region.

Through open dialogue and a collaborative methodology inspired by the PASCAL2 model, low-resourced African languages were identified as a major blind spot. A fellowship was initiated and a set of language dataset challenges established, all to incentivise the creation, collation and uncovering of African language datasets. This five-month process saw submission of 35 datasets from a variety of African languages/dialects with more than 190 data scientists enrolled to solve the challenges. The resulting datasets were released to the African crowdsourcing machine-learning challenge platform Zindi (2022). They were also published on a dedicated channel in Zenodo (2022). The awards given to the community to solve these challenges ranged from USD 500-3 000 each.

Held during the first COVID-19 lockdown in 2020, this data challenge attempted to incentivise the African AI community to engage in the global response to the pandemic. Data scientists were asked on Zindi to predict the spread of COVID-19 around the world over the following few months (Zindi, 2022). Solutions were evaluated against subsequently collected data. This challenge contributed to the global body of knowledge helping stem the impact of pandemics of different sorts. The top three solutions were made available on GitHub (2022). In all, 773 data scientists enrolled in the challenge, leading to 777 submissions. For the winning entry, the average estimate of daily cumulative deaths per country was within 208 of the actual number. The selected projects were awarded between USD 500-1 000 each.

The cumulative efforts of these tailored small-scale investments resulted in significant research initiatives on different scientific and non-scientific, engineering and educational topics. They had particular successes in profiling African languages. Evidence of success has helped unlock several other major funding initiatives. IDRC, for example, extended the AI4D initiative and created the Lacuna Fund, which mobilises funding for labelled datasets that solve urgent problems in low- and middle-income countries (Lacuna, 2022).

The outcomes described below illustrate that small-scale initiatives can be effective in helping build AI capacity, for science and other purposes, in low-income countries:

  • Empowerment of grantees by creating a level playing field where researchers and data practitioners in developing countries were trusted and treated equally to those in developed countries. This considered their financial and operational constraints, and involved them in shaping the funding agenda. The principle of intentionally avoiding any biases and instead creating a machine-learner to machine-learner relationship and working environment proved highly beneficial.

  • International recognition of the creation of the African languages’ datasets in 2021 when two of the outcomes received the Wikimedia Foundation Research Award of the Year (Wikimedia, 2022). Neketo et al. (2020) were awarded for the paper “Participatory research for low-resourced machine translation: A case study in African languages”. The development of the Masakhane online community of AI practitioners was also awarded. This community has attempted to fundamentally change how to approach the problem of under-resourced languages in Africa. Both the work of the authors and the community have been recognised across Africa.

  • Direct support from K4A to establish the Masakhane Research Foundation in Nairobi and the Tanzania AI Community in 2022. This support will potentially empower a bottom-up community of researchers and practitioners to transform itself from an informal group into a registered legal entity.

  • The readiness of a number of donors to create AI-specific funding schemes. For example, IDRC and the Swedish International Development Cooperation Agency launched a four-year CAD 20 million partnership, beginning in 2020, to address a range of AI challenges in Africa.

  •, the Rockefeller Foundation, IDRC and Germany’s Development Cooperation agency making combined contributions to the Lacuna Fund of several million US dollars. This support came to fruition thanks to the portfolio of tangible micro-projects making the case for such a large investment.

  • A gender-diverse distribution of participants. A mixed gender managerial model, along with joint male and female project leads, helped create a research environment more favourable to female principal investigators.

The budget for all the described micro-activities was a modest CAD 500 000 over three years. This modest funding has nevertheless created significant results. This outcome is in some ways comparable, but in a different context, to the large and long-term collaborations (around 20 years) in the European networks of PASCAL, PASCAL2, ELISE and Humane AI. These networks were made possible by relatively modest funding of approximately EUR 50 million. In all cases, the micro-project model has been effective in fostering long-term innovation and impact through small-scale direct funding with minimal bureaucratic overhead.

Financial and managerial tensions arise when creating large-scale projects with ambitious, exploratory or ground-breaking visions that lack the assurance of near-term profitability or benefit, or a clear strategy on how to achieve those goals. In such cases, it is difficult to maintain research cohesion and keep a focus on creative ideas in the ways that smaller initiatives can. The much-criticised Human Brain Project (Enserink and Kupperschmidt, 2014) is perhaps an example of the former. By contrast, an example of the latter is the PASCAL Visual Object Classes challenge (PASCAL, 2022). This challenge is still relevant after 17 years, providing the vision and machine-learning communities with a standard dataset of images and annotations, and standard evaluation procedures.

Micro-projects can create dispersed rather than centralised impacts. However, co-ordinating micro-projects as part of a larger coherent programme might deliver the best of both worlds. PASCAL2 used a bottom-up and small-scale agile funding structure but around a co-ordinated research and collaborative theme of pattern analysis and machine learning. The Humane AI network is undertaking a similar experiment with its micro-project funding programme. Again, it is co-ordinated around a series of themes and “grand challenges”. This network has yielded some promising initial results, delivering almost 60 micro-projects in two years. The answer appears to be “yes, we can”. However, the key is good funding management rather than funding scale or a central intellectual authority that directs the research.

On first impression, independently of the funding mechanism, there is a case for sub-Saharan Africa to receive much greater funding than that available to K4A. However, funding should be disbursed in a way that enables researchers maximum opportunity to unleash their potential to innovate. This can accelerate innovation, the production of cutting-edge research accepted at top AI conferences and establishment of trust with international research institutions and donors. Such developments could go some way to closing the divide between developed and developing countries in terms of scientific achievements, enabled in part by AI tools and access to AI-related education through the Internet.


Aczel, B, B. Szaszi and A.O. Holcombe (2021), “A billion-dollar donation: estimating the cost of researchers’ time spent on peer review”, Research Integrity and Peer Review, Vol. 6/14,

AI4D (2022), Artificial Intelligence for Development Africa website, (accessed 10 August 2022).

Butcher, N. et al. (2021), “Artificial intelligence in sub-Saharan Africa, 2021”, International Research Centre in Artificial Intelligence under the auspices of UNESCO,

CORDIS (2022), “Pattern analysis, statistical modelling and computational learning 2” (fact sheet), CORDIS, European Commission,

DSA (2022), “African AI Research Award 2022”, webpage, (accessed 11 September 2022).

DSN (2022), Data Science Nigeria website, (accessed 11 September 2022).

Deep Learning Indaba (2022), “Deep Learning Indaba 2022”, webpage, (accessed 16 August 2022).

ELISE (2022), European Network of AI Excellence Centres website, (accessed 5 July 2022).

Enserink, M. and K. Kupperschmidt (2014), “Updated: European neuroscientists revolt against the E.U.'s Human Brain Project”, 11 July, Science,

GitHub (2022), “Zindi wins AI4D Predict the Global Spread of COVID-19 Insights”, webpage, (accessed 14 August 2022).

Gwagwa, A. et al. (2021), Responsible Artificial Intelligence in sub-Saharan Africa: Landscape and General State of Play, International Research Centre in Artificial Intelligence under the auspices of UNESCO,

Humane AI Net (2022), European Network of Human-centered artificial intelligence website (accessed 20 September 2022).

K4A (2022a), Knowledge for All website, (accessed 20 September 2022).

K4A (2022b), “Emerging Economies Artificial Intelligence Ecosystem Directory”, webpage, (accessed 21 September 2022).

Lacuna Fund (2022), Lacuna Fund website, (accessed 5 July 2022).

Masakhane (2022), Masakhane website, (accessed 16 June 2022).

Naixus (2022), Naixus website, (accessed 11 November 2022).

Neketo, W. et al. (2020), “Participatory research for low-resourced machine translation: A case study in African languages”, arXiv, arXiv:2010.02353 [cs.CL],

PASCAL (2022), The PASCAL Visual Object Classes Homepage website, (accessed 20 August 2022).

UN (2022), “Sustainable Development Goal 6”, webpage, (accessed 2 June 2022).

VideoLectures (2022), VideoLectures website, (accessed 15 July 2022).

Wikimedia (2022), “Wikimedia Foundation Research Award of the Year”, webpage, (accessed 18 August 2022).

Zenodo (2022), “African Natural Language Processing (AfricaNLP)”, webpage, (accessed 18 August 2022).

Zindi Africa (2022a), “GIZ AI4D Africa Language Challenge – Round 2”, webpage (accessed 22 June 2022).

Zindi Africa (2022b), “AI4D Predict the Global Spread of COVID-19”, webpage, (accessed 22 June 2022).


← 1. Burkina Faso (Building a Medicinal Plant Database for Preserving Ethnopharmacological Knowledge in the Sahel), Burkina Faso (Preservation of Indigenous Languages), Kenya (A Public Dataset on Poaching Trends in Kenya and a Study on the Predictive Modeling of Poaching Attacks), Kenya (Early detection of preclampsia using ambulatory blood pressure monitoring using wearable devices and Long Short Term Memory Networks (LSTM-NN) on the edge), Malawi (A Semi-Automatic Tool for Meta-data extraction from Malawi Court Judgments), Morocco (Arabic Speech-to-MSL Translator: Learning for Deaf), Nigeria (Using Artificial Intelligence to Digitize Parliamentary Bills in Sub-Saharan Africa), Tanzania (A Computer vision Tomato Pest Assessment and Prediction tool), Tanzania (Effective Creation of Ground Truth Data-set for Malaria Diagnosis Using Deep Learning), Tanzania (Improving the Pharmacovigilance system using Natural Language Processing on Electronic Medical Records).

Metadata, Legal and Rights

This document, as well as any data and map included herein, are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area. Extracts from publications may be subject to additional disclaimers, which are set out in the complete version of the publication, available at the link provided.

© OECD 2023

The use of this work, whether digital or print, is governed by the Terms and Conditions to be found at