By now, the global discourse on artificial intelligence has moved far beyond the realm of science fiction. AI is no longer a futuristic abstraction. It is shaping economies, governance systems, education, medicine, defence, banking, media, and even the democratic process itself. Across the world, nations are racing to develop national AI frameworks and policies that will guide how this powerful technology is deployed, regulated, and governed.
It is therefore both significant and instructive that the Government of South Africa recently withdrew its proposed national artificial intelligence framework after concerns emerged regarding questionable references and citations contained in the document, references reportedly suspected to have been inaccurately generated through AI tools.
What happened in South Africa is not merely a policy embarrassment. It is, in my respectful view, a defining moment in the global conversation around the ethical and responsible use of AI.
More importantly, it is also a remarkable demonstration of leadership, humility, accountability, and institutional honesty.
The South African government deserves commendation, not condemnation, for acknowledging the error and withdrawing the document for reassessment by an independent panel of experts.
In an age where many governments, institutions, corporations, and even scholars often double down in defence of obvious mistakes, the willingness to publicly admit shortcomings and recalibrate direction is one of the highest forms of leadership maturity.
That decision reflects a principle I have consistently advocated in my lectures, writings, boardroom engagements, and policy conversations across Africa and internationally: the urgent necessity for Responsible-Human-in-The-Loop (RHITL) systems in artificial intelligence deployment.
For years, many AI scholars and practitioners have rightly spoken about Human-in-the-Loop (HITL) frameworks, systems where human beings supervise, validate, and guide AI outputs. However, I have consistently argued that mere human involvement is insufficient. The real requirement is Responsible-Human-in-The-Loop.
The reason for my addition of “responsible” to the hitherto-conceived human-in-the-loop is the fact that an irresponsible human supervising AI can be far more dangerous than the AI itself.
Artificial intelligence, particularly generative AI systems, can produce extraordinary outputs within seconds. It can summarise books, draft legal documents, generate computer code, create policy papers, produce videos, and even simulate academic references with astonishing fluency. Yet AI systems do not possess wisdom, conscience, historical sensitivity, or moral accountability. They predict patterns based on data. They do not “understand” truth in the philosophical or ethical sense that human beings do.
This distinction is critical.
One of the greatest dangers confronting governments, universities, media institutions, and corporations today is the temptation to confuse linguistic sophistication with factual accuracy.
An AI-generated sentence may sound intelligent while being fundamentally false. An AI-generated citation may appear academically rigorous while referring to a non-existent source.
An AI-generated policy recommendation may look persuasive while containing deeply flawed assumptions.
This is precisely why governance frameworks around AI must themselves be subjected to rigorous human verification, multidisciplinary scrutiny, and ethical oversight.
Ironically, the South African AI framework controversy has now become a powerful real-world case study demonstrating why Africa needs stronger AI governance structures, not weaker ones.
It also underscores another important reality: Africa cannot afford superficial engagement with artificial intelligence. This is part of the assignments that fall within the purview of the President Paul Kagame-led Smart Africa, Africa AI Council initiative.
We must resist the dangerous temptation to merely “import” AI narratives, frameworks, and technologies without building deep indigenous intellectual capacity around them.
Africa’s relationship with technology has historically been shaped by dependency. During earlier industrial revolutions, the continent largely became a consumer rather than a creator. We must not repeat that error in the Fourth Industrial Revolution.
Artificial intelligence is too consequential.
The future of national security, education, economic competitiveness, healthcare delivery, judicial systems, financial services, and even cultural identity may increasingly depend on how nations govern AI.
That is why policy frameworks must be built on intellectual rigour, verified scholarship, transparency, and contextual understanding.
To its credit, South Africa has shown institutional maturity by stepping back and choosing reassessment over denial.
That decision ought to ultimately strengthen, rather than weaken, the credibility of its eventual AI framework.
Indeed, there is a larger lesson here for governments across Africa. And that is the realisation that leadership is not the absence of mistakes.
Leadership is the courage to acknowledge mistakes and correct them before they become institutional disasters. President Cyril Ramaphosa must be commended for this bold action.
In truth, AI hallucinations and fabricated references are now a globally recognised challenge. Even some of the world’s leading law firms, academic institutions, journalists, and researchers have faced controversies arising from AI-generated inaccuracies.
This is not exclusively a South African problem. It is a global AI governance problem. The difference lies in how institutions respond when errors are discovered.
The South African response demonstrates that transparency and corrective action are still possible in public governance.
As African nations increasingly develop AI strategies and national digital transformation agendas, there are several lessons we must urgently internalise.
First, AI policy development must be multidisciplinary. Technologists alone cannot design national AI frameworks. We need ethicists, historians, lawyers, philosophers, sociologists, economists, educators, cybersecurity experts, linguists, and governance professionals at the table.
Secondly, governments must establish robust verification mechanisms for all AI-assisted outputs, especially official policy documents.
Thirdly, African countries must invest heavily in AI literacy, not only among software engineers but also among policymakers, judges, legislators, regulators, journalists, educators, and corporate boards.
Lastly, we must avoid both extremes: blind AI enthusiasm and irrational AI fear.
Artificial intelligence is neither a god nor a demon. It is a powerful tool.
Like electricity, nuclear energy, aviation, or the internet, its impact depends largely on the values, competence, and intentions of those deploying it.
This is why I continue to insist that the future belongs not merely to AI-enabled societies but to responsibly AI-governed societies.
Africa must therefore pursue AI development anchored on ethics, accountability, transparency, human dignity, inclusion, and cultural sensitivity.
The continent possesses immense intellectual capital, youthful demographics, entrepreneurial energy, and untapped innovation potential. But those strengths must be matched with institutional seriousness and governance discipline.
In many ways, the South African episode may become one of the most important AI governance lessons Africa has yet experienced. Not because an error occurred. But because leadership chose honesty over ego.
Correction over concealment.
Accountability over defensiveness.
As artificial intelligence continues reshaping the world, one truth is becoming increasingly clear:
The greatest risk may not be artificial intelligence itself.
The greatest risk may be irresponsible human intelligence supervising it.
And that is precisely why the future of AI must remain firmly anchored in responsible human-in-the-loop systems.
Note: Sonny Iroche is an Oxford-trained AI scholar and a member of the following AI committees: Nigerian National AI Strategy, UNESCO TWG on AI Readiness Assessment Methodology, and Africa AI Council TWG.
