AI needs the foundations the leapfrog story said we could skip
United Nations research into AI in developing-country government shows the technology accelerates what already works and exposes what does not.
Sign up for my Substack daily AI newsletter here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
The leapfrog story has been one of the most comforting narratives in African technology policy for twenty years. The idea is that developing countries can skip the heavy lifting of institutional reform by jumping straight to the latest technology, as mobile phones once leapfrogged fixed-line telecoms. Artificial intelligence is now being sold back to governments in Pretoria, Nairobi and Abuja with the same framing, and recent United Nations University research on public-sector AI in developing countries has quietly undermined it. The finding is that AI accelerates what already works in a government and exposes what does not, and that administrations with weak data, weak procurement and weak accountability do not leap; they stumble faster. For a South African government preparing a national AI policy, that distinction is the whole conversation.
CONTEXT AND BACKGROUND
The United Nations work on AI in public administration sits inside a broader picture that is now reasonably clear. The OECD recorded in April that African AI governance remains in transition, with thirteen national strategies adopted since 2018 and a Continental AI Strategy endorsed by the African Union, but with implementation lagging the strategic ambition and the data protection, procurement and audit systems that would make the strategies operable not yet in place. The World Bank, in a comment that reads as both encouragement and warning, called AI a lifeline for developing economies while making clear that the lifeline requires reliable electricity, broadband and skills on the receiving end.
The Africa AI Policy Lab and Lawyers Hub, in their Africa AI Governance Index, have tracked roughly fifteen African states with national AI frameworks adopted and another twenty-four drafting them, with every sub-region scoring highest on strategy and vision and lowest on implementation and impact, so that the binding constraint has shifted from vision to delivery. Strategy without foundation is becoming the signature failure mode of this moment.
INSIGHT AND ANALYSIS
Three threads run through the United Nations’ work and the surrounding literature. The first is that AI amplifies the state as it is, not the state as the brochure describes it. A revenue service with clean data and strong internal audit will use AI to find tax leakage. A revenue service without those things will use AI to automate its existing errors at scale. The second is that governance mapping is now the binding constraint. CircleID has argued that African AI adoption is running ahead of the risk regimes that should accompany it, with data protection authorities under-resourced for the volume of automated decision-making already in production. The third is a procurement thread. Intelligent CIO Africa reported in July that the Gauteng Department of e-Government has been allocated R1.9 billion for the 2026/27 financial year to drive digital transformation and to stand up a new Artificial Intelligence Office to coordinate responsible AI adoption across the provincial government, a scale of expenditure the Auditor-General and the Standing Committee on Public Accounts have no established methodology to interrogate as algorithmic rather than financial spending.
Traditional auditing asks whether an invoice matches a good delivered. Algorithmic auditing has to ask whether training data, model weighting and automated decisions comply with the Promotion of Administrative Justice Act and the Promotion of Access to Information Act. South Africa does not yet have that capability sitting anywhere.
IMPLICATIONS
For South Africa, the specific implication is that the National AI Policy, now moving through the Department of Communications and Digital Technologies and the NACI advisory process, cannot be written as if the foundations are in place. The policy has to commission the foundations as its first act. That means expanding the Information Regulator’s mandate from POPIA data protection into algorithmic accountability, with staff who can read a model card and the authority to act on what they find. It means requiring every government AI procurement over a prescribed value to publish a model risk assessment, giving the Auditor-General a mandate to audit algorithmic systems the way it audits financial systems, and placing a plain-language public register of government AI in the public domain.
The policy choice in front of South Africa is whether its AI document reads like a technology brochure or like a procurement rule book. Nigeria is already moving faster than South Africa on parts of this, with NITDA framing AI as strategic infrastructure that must cover connectivity, compute, data centres, cloud, data governance and the models themselves, a National Digital Cloud Policy targeting roughly $750 million in private investment, and a 3MTT skills programme that has already registered over 1.8 million Nigerians, while the gap keeps widening until South Africa’s policy is costed and signed.
CLOSING TAKEAWAY
The leapfrog story was always a flattering distraction. It told African governments they could buy their way past the slow work of building capable institutions. Artificial intelligence has made that illusion untenable, because it is a technology that only compounds what you already have. A South African AI policy that pours capital into deployment while leaving data quality, audit capacity and the Information Regulator untouched will accelerate the state we already have, not the state we have promised our children. The foundations are not optional, and they are not glamorous. They are the policy.
Author Bio
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments