Globally, tackling the digital divide is an important development priority. Approximately 2.6 billion people – a third of the population - remain offline. Women are 19% less likely than men to use mobile internet. Digital public services still benefit mostly young, urban, higher-income, and digitally savvy men. This exclusionary trend has far-reaching economic and social implications, with 32 lower-income countries alone losing around US$1 trillion in tax revenue, productivity, and other benefits due to the digital gender divide. While some of these digital gaps are closing, a new inequity is emerging; Compute power, for artificial intelligence and the ability of emerging countries to leverage technological developments as a key accelerator of attaining their sustainable development goals. Compute power is fast becoming the new and the next facet of the digital divide. 
 

1. A new era of opportunity – and inequality

Compute describes the computational power and technology available to train and refine artificial intelligence models and processes. Shifts in how AI models are developing, especially advances in research and development in hardware usage and significant investment in custom hardware and technology have potentially transformed the AI landscape. The ‘modern era’ of how AI systems are trained represents the potential divergence between higher-income and lower-income countries.

In a critical year of the Global Digital Compact, an initiative to ensure that digital technologies work for the benefit of all, UNDP is working to bridge the local and the global – driving action and meaningful change in the areas of funding, data, talent, regulation - and compute power. Recent discussions across Africa have highlighted how today’s opportunities around AI need collective actions despite there being different development statuses between countries. Collective action is a fundamental driver of the UNDP-G7 industry partnership to explore developing the AI Hub for Sustainable Development with a particular focus on Africa.

This direction is shaped by the important recognition that the compute and AI digital divide is both a new and important inequity but is also founded on and catalyzed by existing deeper and pervasive divides. Global South ecosystems that are already digitally constrained do not yet have the capacity and resources to research and develop at the pace and concentration that richer countries can. Existing barriers faced by women and girls pursuing STEM careers in such countries lead to a lack of available AI and digital talent. Challenges in resourcing data collection, curation, and management prevent public and private sector innovators from exploring the power and potential of AI to address local challenges and priorities. 
 

2. Closing the gap

Recent discussions on the African continent have highlighted three important priorities in relation to compute, data, and talent.

On democratizing compute

Computing resources across the world are unequally distributed and not accessible to all. Less than 20% of developing countries have modern data infrastructure such as co-location data centers and direct access to cloud computing. Undertaking large scale inference on large scale models requires robust computing power that is expensive. This restricts the local market’s ability to build artificial intelligence that supports policymaking, data-driven decisions, and business opportunities. Computing power is a major challenge for students working on AI projects - some must leave laptops on for several hours or days to enable them to process AI models. There are growing initiatives to tackle compute power for local contexts. One example is People + AI a collective of researchers, companies, start-ups, and non-profits to solve ecosystem-level challenges in India. People + AI through the Open Cloud Compute initiative was recently able to increase access to cost-effective and resilient cloud computing through a digital public infrastructure approach of open networks for compute infrastructure. 
 

On data and data models

The critical lack of datasets that reflect the needs and capacity of the global majority countries is a significant constraint to digital development.  While some data for AI models is available through open data repositories, high-quality, locally relevant datasets in developing countries are often limited and expensive to collect. Researchers need funding to undertake new and inexpensive methods of data collection, curation, analysis, and usage relevant to unique local contexts. Private sector companies are creating open and publicly available datasets to spur research and innovation in Africa. A good example Is Mozilla Common Voice and Fair forward that are open source collectives strengthening Natural Language Processing in the African context.
 

On human capacity and talent

There is a continued need to support the development of science, technology, and innovation pathways, deepen Research and Development cultures, and enhance translational and commercialization skills. Initiatives such as Data Science Africa are shaping AI researcher communities in Africa. GIZ (Gesellschaft für Internationale Zusammenarbeit) is working with Kwame Nkrumah University of Science and Technology ( KNUST) Responsible AI Lab to develop graduate courses in AI for students in Ghana. KNUST has trained more than 200 people in cutting-edge machine learning techniques and data science applications to bolster the capacity to create AI-related solutions fit for the African context.  
 

3. A new development approach is needed

To address the challenges of democratizing compute, strengthening data infrastructure, and fostering sustainable talent pipelines, a new multi-stakeholder partnership is essential. The AI Hub for Sustainable Development, championed by the Italian G7 Presidency with UNDP, aims to a multi-stakeholder initiative to orchestrate collective actions to strengthen local AI ecosystems in developing countries, with a focus on the African continent.  

To do so, the Italian Presidency, together with its knowledge partner, UNDP, proposes that the Hub operates as an AI “ecosystem partnerships and exchange space,” to solve the ecosystem level challenges. Not solving these challenges would lead to a slower pace of innovation, create waste and inefficiency and create unequal and unfair access to AI. This requires working together with the private, public, and non-profit stakeholders to innovate, demonstrate catalytic and scalable actions across three ecosystem level challenges:  

1. Infrastructure challenges: Data gaps - unequal distribution or access in AI models, which is likely to cause data gaps with language; high computing costs slowing innovation without innovative or new strategies to bring down the cost of computation at an ecosystem level and increase access.
2. Talent challenges: Talent in research, exploration of AI use-cases not scaling beyond a few donor-funded initiatives.
3. Policy challenges: Unclear robust and innovative AI policies that are not leading to local ecosystem growth and leaving gaps in digital safeguards for people. 

As the Italian led G7 and UNDP co-designs the AI Hub for Sustainable Development with partners from the global, along with regional, and local industry, it is imperative to build alliances with in-country stakeholders deeply immersed in exploring AI opportunities. The co-design effort must gear towards orchestrating global action to benefit local AI ecosystems.  Meanwhile, across Africa, stakeholders must continue to convene and act towards leading, learning, investing, and collaborating to harness the potential of AI for Africa’s development.  

Contributors to the discussion during Roundtables in Ghana and Nigeria

Ghana AI Roundtable: (Patricia Poku (Data Protection Commissioner), Kofi Dadzie (Tony Blair Institute), Nii Longdon, Gifty Buah (University of Ghana) Elikplim Sabblah (GIZ), Daniella Darlington (Copianto AI), Blaise Bayuo (ACET), Alhassan Baba Muniru, Daniel Otto, (GFA Consulting) Worlali Senyo, Jeffrey Amasa, (Farmerline) George Arthur-Sarpong (Viamo), Joseph Berkoh (AU Development Agency – NEPAD)

Italy G7: Eva Spina (G7 Chair and Head of Department at MIMIT) Vincenzo Del Monaco (G7 Co-chair and Minister Plenipotentiary), Valeria Vinci (HoU at MIMIT), Eleonora Iannuzzi (HoU at MIMIT)  

Nigeria AI4D: Ada Irkefe, Ayomido Owoyemi, Nonye Ujam (Microsoft), Ojoma Ochai, Olayinka David-West, Sanusi Ismaila, Surryyah Ahmad, Toyosi Arkele-Ongunsiji, Victor Famubode, Dr Ifeoma Nwafor, Nasir Yamama.

UNDP: Keyzom Ngodup Massally,  Calum Handforth, Alena Klatte, Giulia Vitagliano, Alex Hradecky, Oluwatoyin-Samuel Bamidele, Seth Akumani, William Tsuma and Ngasuma Kanyeka