Table of Contents
The AI literacy training for civil servants recently delivered in Nigeria and Ghana reflected the commitment of UNESCO and the respective governments to supporting the ethical adoption of AI in the public sector. The programme helped participants move from limited understanding and uncertainty about AI to a more informed and responsible approach to its adoption. This piece reflects on what I learnt from delivering the training, including the successes we can expect and the new approaches we might adopt to strengthen implementation of UNESCO’s Recommendation on the Ethics of AI (UNESCO Recommendation) and responsible AI governance more broadly. One critical insight is that cultural context matters for understanding and evaluating ethical risks. This led me to consider how an African worldview, and specifically the Yoruba concept of Omoluabi, can help explain and operationalise the UNESCO Recommendation while addressing persistent challenges in implementing AI governance through a culturally sensitive lens.
Cultural context matters
Participants in the AI literacy course often expressed disappointment that the training did not focus on recommending or demonstrating AI tools. This quickly gave way to an appreciation of its broader value: as AI reshapes public service delivery and government interactions with citizens, technical competence alone is not enough. Participants became more open to exploring AI’s opportunities and risks, separating the hype from the reality, and understanding how the training equipped them to evaluate both.
However, one thing quickly became clear: for ethical risks and governance frameworks to resonate, they must be framed in a relatable cultural context. Storytelling and case studies proved invaluable in bringing these concepts to life. For example, I introduced ethical risks, opportunities for AI adoption, and governance frameworks through a story I called “My Village People.” For context, the expression “my village people,” is widely used in Nigeria and Ghana as a metaphor to describe ancestors, family members, relatives, or neighbours who are jokingly blamed for misfortunes or setbacks. While rooted in cultural beliefs about unseen influences, it is now commonly used humorously in everyday conversation and on social media to explain inconveniences, often without any literal belief in supernatural causation.
My Village People is a short story about a team of researchers collecting voice data to develop an automatic speech recognition (ASR) system aimed at advancing financial and language inclusion. However, the communities they hoped to benefit resisted participating, fearing that they might “lose their voices” during the data collection process and have no way of finding the researchers to recover them. While these concerns may be easily dismissed as superstition, the story reveals deeper ethical issues that are central to UNESCO Recommendation including privacy and data governance (the fear of “losing their voices” in particular clearly maps onto concepts of control, consent, data governance and trust), inclusion, linguistic diversity, digital divide, and the importance of meaningful stakeholder engagement. By grounding these
principles in a familiar African context, the story makes abstract ethical concepts tangible and demonstrates why the responsible and ethical design of AI must take indigenous values and lived experiences seriously.
Liability remains critical for public sector AI adoption
The story-based exploration of ethical risks may have increased interest in identifying and evaluating opportunities for public sector AI adoption, but this enthusiasm is matched by concerns about liability. Participants’ comments and questions revealed broader concerns about how public sector actors will be protected from liability given ongoing challenges with AI trustworthiness. They also questioned whether emerging governance frameworks are sufficiently adaptable to African contexts and realities. For example, would early regulation stifle innovation, undermine the public welfare objectives driving AI adoption, or leave Africa behind? Would courts be inundated with ethically complex cases on liability for AI-related harms?
Presenting the broad consultations that informed the development of the UNESCO Recommendation and its subsequent adoption by 193 Member States helped address some of these concerns. The Recommendation demonstrates that ethical standards cannot be sustainably developed through the lens of a single culture or civilisation. Towards the end of the training, however, I realised that grounding these principles in African philosophical perspectives and worldviews, such as the Yoruba concept of omoluabi, would make the framework more accessible and deepen participants’ understanding. But accessibility was not the only gain. Framed this way, omoluabi did more than illustrate the Recommendation, it spoke directly to the liability and accountability worries raised above, giving public actors a shared language for locating responsibility rather than letting it dissolve into “the algorithm decided.” The rest of this piece develops that claim.
Omoluabi in AI ethics and governance
Omoluabi, like Ubuntu offers a rich conceptual resource for articulating an African worldview of technology governance. In Yoruba moral philosophy, omoluabi embodies the notion of the ideal person and forms the core focus and goal of Yoruba traditional education. Broadly understood as signalling a person of good character, omoluabi literally means Omo ti olu iwa bi, eni ti a ko ti o si gba eko, meaning a person who is properly and well nurtured and who lives by the precepts of the education s/he has received. While omoluabi is usually associated with human character formation, its logic can be extended to non-human entities. A company that prioritises the rights, interests and needs of the communities in which it operates is an omoluabi company. An omoluabi government demonstrates transparency in public spending and prioritises welfare and respect for the rule of law.
Applied to AI, omoluabi works at two levels. First, and most importantly, it is a standard for the actors – the companies that build these systems and the governments that buy and deploy them; the omoluabi company and omoluabi government just described are the primary bearers of character here. Second, it offers a diagnostic lens on the systems themselves: a way of gauging how far a system’s behaviour falls short of what an omoluabi would do, and whether it embodies the core pillars of responsible AI – inclusivity, accountability, transparency, safety and explainability. Character, strictly speaking, belongs to the actors; speaking of a system’s ‘moral character’ is shorthand for this diagnostic test, since the system has none of its own.
For example, whereas omoluabi is a moral character cultivated through lived experience and socialisation, AI’s amoral nature and potential for harm undermines any claim that it behaves according to the precepts of its training, or that there are training precepts at all. Omoluabi abhor lies, unjustified selective treatment, sycophancy and feigned knowledge. In contrast, AI hallucinations are rendered with the same ‘confidence’ it uses for ground truths. LLMs routinely default to gender stereotypes, which in omoluabi terms is worse than silence. An omoluabi who does not know or has no answer to a question is expected to defer or stay quiet, rather than render nonsense (ìsọkúsọ) which undermines his/her character. Omoluabi also values wisdom and restraint. Not everything that can be done should necessarily be done. This principle is highly relevant in an era of rapid AI experimentation and competitive technological acceleration. Current AI development is often driven by speed, scale, and disruption. Companies race to deploy increasingly powerful systems without fully understanding their societal consequences. This culture conflicts sharply with the omoluabi emphasis on thoughtful and responsible conduct. An omoluabi approach would encourage caution in high-risk AI deployment and support participatory innovation processes that prioritize societal welfare.
Finally, the omoluabi ideal condemns irresponsibility and lack of accountability, while AI systems face no consequences and cannot be shamed or rewarded into modifying their behaviours to align with omoluabi values. Precisely because the system cannot be held to account, the demand falls back on the actors who chose to build and deploy it. Harm caused by AI systems is often obscured through claims that outcomes are technically or algorithmically determined.
From a liability and broad governance perspective, the omoluabi framework rejects the moral distancing that ecosystem actors appear to struggle with, and the vacuum in accountability that this creates. For civil servants, this is not only an ethical stance but a practical answer to the liability worries above. For example, an omoluabi approach turns the values into procurement discipline: it presses buyers to take a questioning stance towards vendor claims, to require explainability and documented evidence that a system defers when it does not know, and to fix clear lines of human responsibility before any high-risk deployment. Locating responsibility this way does not expose public sector actors to more liability but rather protects them, because it stops the accountability vacuum forming in the first place. It also guards against importing extractive growth models and reproducing inequality.
READ MORE: https://www.unesco.org/ethics-ai/en/articles/we-can-re-imagine-ai-ethics-through-yoruba-concept-omoluabi?hub=701
