AI recommends; humans decide: a governance imperative for African institutions

Karikari Achireko Kwagyan, Justice Ohene-Akoto, Adzasey John Yaw

Oct 25 2026 · 11 min read

AI · Corporate Governance · Public Governance

The accountability question every African leader must be able to answer

When an AI system makes a recommendation that leads to harm, or when it rejects a loan application, flags an individual as a fraud risk, or denies access to a public service, who is held accountable?

This is not a hypothetical scenario. AI systems are already making consequential recommendations across virtually every sector of modern economic life, and the humans who deploy them remain legally, ethically, and institutionally responsible for the outcomes. The assumption that accountability can be delegated to the technology or to the vendor, or to the technical team, is one of the most dangerous governance failures of our time.

This article is adapted from the Africa Sustainable Energy Centre (ASEC) Board AI Awareness Workshop, led by digital literacy and data privacy expert Kate Watson. It seeks to translate the concept of accountability into practical guidance for governance. Rather than targeting AI developers, it is intended for board members, executives, and institutional leaders who carry decision-making and oversight responsibilities in an increasingly AI-driven environment. At its core is a clear and fundamental message: while AI can provide recommendations, the responsibility for decisions remains with humans.

What is actually AI, and why are the misconceptions dangerous?

Artificial Intelligence (AI) is often misunderstood. Some people see it as a perfect system that always gets things right, while others view it as a futuristic technology that has little relevance to today’s work. Both views can create risks for organisations and those responsible for governance.

A helpful way to think about AI is through the iceberg analogy. What people see on the surface, such as a chatbot, an automated recommendation, or a risk score, is only a small part of the system. Beneath the surface are the data used to train the AI, the decisions made by its designers, and the assumptions, patterns, and potential biases built into that data. To govern AI effectively, leaders need to understand not only what the system produces but also how those outputs are generated.

At its foundation, AI is a tool that uses data to identify patterns and make predictions or recommendations. It does not think, reason, or understand situations as humans do. Instead, it generates responses based on what is statistically most likely. As a result, AI outputs can sometimes be inaccurate, biased, or produce unintended consequences if they are not properly reviewed.

The table below introduces key AI concepts that board members, executives, and governance professionals should understand to make informed decisions about the use of AI in their organisations.

Term	Governance-relevant definition
Machine learning	Systems that learn patterns from data rather than being explicitly programmed. The output quality depends entirely on the quality and representativeness of training data.
Large language models (LLMs)	AI trained on vast text data to predict and generate language. This includes ChatGPT, Gemini and Copilot, among others. They predict likely word sequences; they do not verify truth.
Generative AI	AI that creates new content (text, images, audio, code or video) based on learned patterns or trained models. The most publicly visible branch of modern AI.
Agentic AI	AI systems capable of taking autonomous actions toward a goal, moving from passive response to active task execution. Raises significant accountability questions.
AI bias	Systematically unfair outcomes produced by AI systems, typically due to imbalanced training data or flawed design objectives. A core governance concern.
AGI	Artificial General Intelligence: a theoretical form of AI that could match or exceed human intelligence across all domains. Does not currently exist.

Table 1: Core AI concepts for governance literacy.

The acceleration gap: why the governance urgency is justified

The pace of AI capability development is without precedent in the history of technology. Within one year of AI development in terms of capability advancement, it is roughly equivalent to three to five calendar years of progress in most other technology domains. Tools that seemed experimental twelve months ago are now embedded in enterprise workflows, public services, and consumer products across the world.

In all these developments, what is not getting the attention is governance maturity. There is an estimated one-to-two-year governance lag between the capabilities that AI systems now possess and the institutional frameworks, board competencies, and accountability structures required to govern them responsibly. This is not a technical inconvenience but a strategic and institutional risk. Each capability advance that outpaces governance is a consequence waiting to happen.

The AI risk landscape: six domains that demand board attention

AI does not create entirely new categories of risk. It amplifies existing ones, increasing their speed, scale, and systemic reach. Six risk domains deserve particular attention from boards and governing bodies:

Risk domain	Description	Governance implication
Inaccuracy	AI generates probabilistic outputs, not verified facts. It can fabricate references, misrepresent context, and present false information with apparent confidence.	AI outputs must support human decisions, not replace them. Verification protocols are essential.
Bias and discrimination	AI can replicate and amplify historical biases embedded in training data, producing systematically unfair outcomes for marginalised groups.	Bias testing is a growing regulatory and governance expectation. Affected communities must be considered.
Over-reliance	Humans may fail to question AI outputs, particularly under time pressure or when recommendations are presented with apparent authority.	Human-in-the-loop oversight, the requirement for humans to review and challenge AI recommendations, is non-negotiable.
Lack of transparency	Many AI systems operate as “black boxes” whose internal logic is not interpretable even to their operators.	If a decision cannot be explained in plain language, it cannot be governed effectively. Explainability is a governance prerequisite.
Data protection	AI systems require large volumes of personal and sensitive data. Informal adoption without governance alignment dramatically increases breach risk.	Data governance and AI governance must be integrated, not treated as separate disciplines.
Reputational damage	The reputational consequences of AI failure can be faster and more severe than regulatory action, as stakeholder expectations of fairness rise.	Trust is a strategic asset. Institutions that cannot demonstrate responsible AI governance risk losing it before regulators act.

Table 2: AI risk domains and their governance implications.

The African dimension: particular risks, distinctive opportunities

Many of these AI risks are especially important for African institutions. Most advanced AI systems today are trained mainly on data from people in Western countries, especially in the Global North. As a result, these systems may not fully understand African cultures, realities, and experiences. This can lead to less accurate results, cultural misunderstandings, or unfair outcomes for African users.

In many African countries, data protection laws and regulations are still developing. This increases the risk that AI tools may be used without proper safeguards, leading to privacy violations or misuse of personal information. Another major concern is misinformation. AI can create content that sounds convincing but is false, which can be particularly harmful in places where reliable information is limited and public trust in institutions is difficult to build.

At the same time, these challenges create important opportunities. Africa has shown before that it can adopt new technologies in innovative ways without following the same path as other regions. Mobile banking is a well-known example. The same approach can be applied to AI governance. By creating strong and responsible AI rules now, before harmful practices become common, African institutions can develop systems that reflect African values, meet the needs of African communities, and set an example for the rest of the continent.

A practical governance framework: five pillars for responsible AI deployment

Effective AI governance is not fundamentally a technology problem, but an institutional design problem. The following five pillars provide a practical framework for organisations seeking to move from informal, ungoverned AI adoption to structured, accountable deployment:

Visibility. Know where AI is being used across your organisation, what decisions it is influencing, and who is affected. An AI inventory, a comprehensive map of AI touchpoints, is the governance baseline from which everything else follows.
Accountability. Define, in writing, who owns AI-assisted decisions and who is responsible when they cause harm. Diffuse accountability is a governance vacuum. It must be replaced with named ownership.
Human oversight. Establish mandatory human review at every high-stakes decision juncture, those affecting access to services, employment, credit, or legal standing. Override authority must be explicit, exercised regularly, and protected from institutional pressure to defer to the machine.
Explainability. Require that any AI recommendation affecting a significant decision can be articulated in plain language to the individual it affects. If it cannot be explained, it cannot be governed or controlled, and it should not be relied upon.
Incident governance. Implement formal protocols for reporting, documenting, analysing, and learning from unexpected AI behaviour. Near-misses are governance intelligence. Organisations that do not capture them are flying blind.

A critical cultural dimension underlies all five pillars: the creation of a challenge culture. Boards bear direct responsibility for building institutional environments in which professionals at all levels feel empowered and expected to question AI recommendations rather than accept them uncritically. Time pressure, commercial incentives, and the perceived authority of data-driven outputs can erode human judgment through disuse. Resisting this tendency is a board-level responsibility.

Vendor accountability: AI governance extends beyond the organisation

AI governance does not begin and end with internal processes. It must extend to the vendors and platforms through which AI capabilities are accessed. When procuring AI tools, organisations should conduct due diligence that goes well beyond functionality and pricing. Key questions include:

What data was the system trained on, and has it been tested for bias and accuracy in African contexts?
Can the system’s recommendations be explained in plain language to the individuals they affect?
What data does the system collect, store, and process, and under what legal basis?
What security certifications and data protection commitments does the vendor hold?
How frequently is the model updated, and what notification will we receive of material changes?

The final question is particularly important. AI systems are not static products. They change, sometimes substantially, as their underlying models are updated. An organisation that assessed a vendor’s system twelve months ago may be operating a materially different system today, without realising it. Governance frameworks must account for this dynamic, not assume a fixed baseline.

Recommendations for African institutional leaders

Based on the ASEC Board AI Awareness Workshop, we advance the following recommendations.

For boards and governing bodies

Conduct an immediate AI inventory: commission a comprehensive assessment of how AI currently influences decisions within your organisation and across your value chain.
Establish board-level AI oversight: designate a committee or named board member with explicit responsibility for AI governance.
Formally adopt the central principle: AI recommends; humans decide, and build institutional structures to enforce it.
Create a challenge culture: establish protocols and cultural expectations that require professionals to question, verify, and override AI recommendations where appropriate.

For executive management

Map all AI touchpoints: document every decision process in which AI plays a role, identifying stakes and current levels of human oversight.
Implement incident governance: establish formal reporting, documentation, and learning protocols for unexpected AI behaviour.
Apply rigorous vendor due diligence: require evidence of bias testing, explainability, and data protection compliance from all AI vendors.
Invest in AI literacy across the leadership team, not only among technology professionals.

For the African policy community

Develop African AI governance standards that reflect continental data realities, cultural values, and human rights commitments.
Demand representational accountability: require that AI systems deployed in African markets demonstrate accuracy and fairness for African populations.
Build regional knowledge networks to facilitate the exchange of governance frameworks and incident learnings across African institutions and jurisdictions.

Conclusion: accountability does not transfer

The overarching message is simple: technology evolves at an accelerating pace, but accountability does not keep pace with it. When an AI system recommends, the human who deploys it remains responsible for the outcome.

This is not a counsel of fear. AI offers genuine and significant advantages in efficiency, analytical power, and the ability to surface patterns that human observation alone would miss, advantages that are particularly relevant for institutions working to address complex challenges in resource-constrained environments. But those advantages are realised only within governance frameworks that protect human judgment, demand explainability, require accountability, and maintain the primacy of human decision-making at the moments that matter most.

The organisations and nations that will lead in the AI era are not those that adopt AI fastest. They are those who adopt it wisest. Building that wisdom through literacy, governance, and a commitment to values that precede and transcend any particular technology is the leadership work of our time.

illuminem Voices is a democratic space presenting the opinions of leading Sustainability Thought Leaders, their views do not necessarily represent those of illuminem.

The world needs sustainability knowledge. At illuminem, no interest group or shareholder can influence our work. Thank you for supporting our mission to make high-quality and independent sustainability information free for all. Every contribution helps. Thank you for donating today.