Artificial intelligence has acquired a familiar vocabulary in the boardroom. Governance, compliance, risk appetite, model inventories, responsible AI, human-in-the-loop. These are necessary conversations, but I fear we are becoming too comfortable with the language of control while avoiding the more difficult question beneath it: what happens to human agency when machines become good enough, cheap enough and trusted enough to make decisions for us? For Africa, that question is not philosophical decoration. It should sit at the centre of our AI strategy because the decisions we make now will determine not merely how we use technology, but how much authority we are prepared to surrender to it.
The continent is approaching artificial intelligence with understandable enthusiasm. AI could transform financial inclusion, healthcare, agriculture, education, public administration and the delivery of services to populations that existing systems have consistently underserved. Governments want investment, banks want efficiency, fintechs want scale and boards want productivity. Everyone has discovered the intoxicating economics of automation: machines do not sleep, demand pensions or complain about repetitive work. Yet efficiency is not neutral. A system that can process ten thousand loan applications in the time a human being considers ten is impressive; a system that can reject ten thousand Africans without any of them understanding why is something altogether different.
This is where the AI governance conversation must become uncomfortable. We have concentrated heavily on whether AI complies with regulation, whether data is protected, whether models are secure and whether organisations can demonstrate governance. Those questions matter enormously, especially as boards become accountable for technologies they may not fully understand. Yet a perfectly governed system can still diminish human beings if its architecture quietly removes their ability to question, contest, understand or influence decisions that affect their lives. Compliance cannot become the moral ceiling of AI governance. Passing an audit does not automatically mean that a system deserves the authority we have given it.
Consider credit. An automated lending model may decide that someone in Lagos, Nairobi or Accra presents an unacceptable risk based on hundreds of variables and correlations. Perhaps the model is statistically excellent. What happens, however, when that individual has no meaningful opportunity to explain an unusual income pattern, challenge incorrect data or ask another human being to reconsider the conclusion? Extend that logic to insurance, employment, university admissions, healthcare, immigration, welfare, policing and access to financial services, and suddenly we are no longer discussing clever software. We are designing the relationship between citizens and institutions, including who gets heard when the machine says no.
There is a particularly African danger here because much of the technology we will deploy will not have been conceived around African social and economic realities. Our economies contain enormous informal sectors. Identity infrastructure differs considerably between countries. Income can be irregular without being illegitimate. Families support one another across borders, addresses can be complicated, employment histories do not always fit neat categories and financial footprints may look unconventional to models trained to recognise something else as normal. If we automate first and interrogate later, we risk converting African difference into algorithmic disadvantage, then congratulating ourselves on the efficiency with which we achieved it.
The answer is not to resist AI. That would be economically foolish and intellectually lazy. Africa should be ambitious about artificial intelligence precisely because it can help us leapfrog broken infrastructure and create services at extraordinary scale. But ambition without governance is merely acceleration, and acceleration is not progress if nobody has checked the direction of travel. Boards therefore need to move beyond asking whether AI can make a particular decision. The more important governance question is whether AI should be permitted to make that decision alone, what happens when it gets the decision wrong and whether the affected human being retains a meaningful route back into the process.
That distinction should become fundamental to African AI governance. For decisions carrying significant consequences for individuals, human participation cannot become ceremonial. A human reviewer who clicks “approve” on ninety algorithmic recommendations before lunch is not meaningful human oversight; neither is an appeals process hidden behind an email address nobody answers. Human involvement must carry actual authority to interrogate the machine, consider context, depart from its recommendation and explain the final decision. Regulators should demand evidence of this, boards should design for it and technology companies should be required to demonstrate it rather than placing “human-in-the-loop” on a governance slide and considering the matter closed.
We also need the courage to establish boundaries. There should be categories of decisions where full automation is unacceptable regardless of how sophisticated the model becomes, not because machines are inherently sinister, but because some decisions carry a social weight that requires accountability capable of understanding context, discretion and consequence. Africa has an opportunity here that mature digital economies did not always have. We are building much of our AI infrastructure while the consequences are still visible ahead of us. We do not have to inherit every assumption embedded in Silicon Valley, Brussels, London or Beijing. Importing technology must not mean importing somebody else’s assumptions about whose circumstances count as normal, what constitutes risk and how much explanation an individual deserves.
The greatest danger, then, is not the cinematic one in which artificial intelligence becomes more intelligent than humanity and takes control. The more immediate danger is quieter: institutions gradually stop asking humans because machines become cheaper to ask. One decision disappears into automation, then another, until eventually nobody remembers that there was once somebody you could speak to, challenge or persuade. Africa should build AI, fund it, teach it and compete globally with it, but we should establish the principle before algorithms become invisible infrastructure: automation may augment human judgement, but efficiency must never become an excuse for engineering human agency out of consequential decisions. Otherwise, we may discover that we built remarkably intelligent systems and, somewhere along the way, designed the human being out of the room.
- business a.m. commits to publishing a diversity of views, opinions and comments. It, therefore, welcomes your reaction to this and any of our articles via email: comment@businessamlive.com
Michael Irene, CIPM, CIPP(E) certification, is a data and information governance practitioner based in London, United Kingdom. He is also a Fellow of Higher Education Academy, UK, and can be reached via moshoke@yahoo.com; twitter: @moshoke
