Governing What We Did Not Build: Africa’s AI Policy Gap

AI is spreading across Africa faster than the institutions tasked with governing it can keep up. Much of the technology driving this expansion, however, is being developed, trained, and tested elsewhere. Dr Edmund Terem Ugar, an AI ethics and governance expert, poses a critical question: how do you govern something you did not create? The fault lines are emerging; for example, South Africa’s draft AI policy was withdrawn after fabricated citations were exposed, while universities have used opaque AI-detection tools to assess student work. Joseph Mugauri, an AI policy and governance researcher, sees the same challenge from inside institutions already deploying AI. African countries, he argues, remain too often positioned as consumers rather than creators of technology. “Our first problem is the mentality of a consumer,” he says. “We need to start seeing ourselves as developers.”

Governance is not ethics

While governance and ethics are often used interchangeably in policy discussions, they serve fundamentally different functions in shaping responsible technology. “Governance is putting the laws in place,” says Dr Ugar. “Ethics asks whether those laws respect human dignity.” The distinction becomes important when dealing with personal data. A customer who clicks ‘accept’ on lengthy terms and conditions may have technically consented to their information being shared or sold. But that does not constitute meaningful consent, says Dr Ugar. “Is it ethical for them to sell your data without compensating you?” Anonymising information does not necessarily remove the ethical concerns. Data represents a person and reveals sensitive details about their life, including finances, gambling habits, mental health struggles, or political interests. The question, ultimately, is whether people who generate valuable data should have meaningful control over how it is used and whether they should share in the value it creates.

When policy cannot define the technology

Mugauri points to another weakness in AI governance: policymakers do not always agree on what they are regulating. Terms such as “explainability”, “interpretability” and “transparency” are frequently used interchangeably, he says, despite describing different concepts. He found similar confusion in South Africa’s withdrawn AI policy, which at different points appeared to treat explainability as part of transparency and interpretability as a route to explainability. If policymakers cannot clearly define these concepts, institutions may struggle to determine what they should require from AI developers and vendors. The problem is not only technical. It is a policy problem. Unclear definitions can produce unclear obligations.

Whose ethics should guide Africa’s AI?

Copying Western ethical ideas into Africa without adjusting them to local realities can create a situation in which a decision that benefits the majority is considered the “right” choice — even if some individuals do not benefit or might be worse off, says Dr Ugar. A rights-based approach, meanwhile, focuses on individual freedom but may overlook the importance of community. “The issue is not that the rights-based ethics should be discarded,” Dr Ugar explains. What is important is ethical compatibility with existing societal norms and context. Therefore, “it is that frameworks built in different historical and cultural contexts can not just be transplanted into African societies without asking what they mean here.” For African policymakers, the challenge is not just how to regulate AI, but also which values should guide that regulation.

Copying foreign laws is not enough

Mugauri says that Africa’s lawmakers are falling into a dangerous pattern of copying regulations from elsewhere without adapting them to local realities. He points to Kenya’s draft AI bill, which borrows heavily from the European Union’s approach. “The risk is that African countries end up with rules designed for very different economies and institutions,” he says The problem starts with a gap between technical and legal knowledge. “The legislators do not understand the technology, and the developers do not understand the law,” Mugauri explains. What is missing is expertise that can bridge the two worlds. In policymaking, who gets invited into these conversations is also a concern. Mugauri notes that the process is often selective, excluding voices that could spot problems before they are written into law. “Building effective AI governance in Africa means not only closing the knowledge gap, but also widening the circle of people involved,” says Mugauri.

Students are already paying the price

The consequences of poorly understood AI are already being felt in universities. While working at the University of Cape Town’s Centre for Innovation in Teaching and Learning, Mugauri encountered students whose work had been flagged by AI-detection software. Some received zero marks. The problem was not simply whether the students had used AI, but whether the institution could explain why the software flagged them. “For as long as you cannot explain why the system flagged AI, you cannot use that system,” Mugauri says. A percentage score, without a clear explanation, should not determine a student's academic future. There is also a question about whether AI detectors can reliably distinguish machine-generated writing from good human writing. Students are trained to follow particular academic conventions, while AI systems are trained on similar patterns. A 2023 Stanford study found that AI detectors misclassified up to 61% of essays by non-native English speakers as AI-generated, compared to near-zero false positives for native speakers. This is a bias with particular implications for African universities, where English is often a second or additional language. Mugauri notes that there is a need for compulsory AI literacy education from the first year of university. Although AI regulation does not eliminate cheating, the objective, instead, should be to prevent students from being punished by systems that cannot explain their own decisions.

From consumers to developers

The broader problem is Africa’s position in the global AI supply chain. Mugauri says many systems deployed on the continent are trained and benchmarked abroad, using datasets that may not adequately represent African populations. That creates risks around privacy, bias and relevance. “We need to do away with the mentality of a consumer,” he says. “We need to start looking at ourselves as developers.” Doing so requires more than encouraging AI startups. It means investing in research, computing capacity, data infrastructure, electricity and technical skills. Dr Ugar points out that differing data protection laws across African countries make it difficult to build systems that operate seamlessly across borders. For investors and developers, that fragmentation increases costs and reduces the attractiveness of the continent as a single market. Greater regional coordination, he says, would help create the scale needed to build African AI capacity.

Build the foundation first

Mugauri is wary of African governments rushing to regulate sophisticated AI while basic digital infrastructure remains unreliable. European regulatory frameworks, he notes, were built on decades of research. African policymakers are often working with far shorter research cycles. “Those countries have been doing research and it goes far back to like 2000s or the 1990s,” he says. “We have six months to a year worth of research and you want to start making decisions based on that.” Africa needs stronger electricity, connectivity, digital infrastructure and technical skills, research institutions, greater computing capacity, policies that retain skilled workers and more coordination between countries. Without that foundation, AI governance risks becoming another exercise in importing solutions rather than building them.