Recent Posts
Imagine a farming cooperative in southern Africa facing another season of uncertain rain. Its members have received a small development grant, but the money is not enough to meet every need. They could repair a shared irrigation channel, buy drought-resistant seed for the poorest households, or help a few productive farms drill private boreholes. Hoping to clarify the options, one member asks an artificial intelligence system how to obtain the greatest return.
The system produces a polished answer. It estimates yields, compares costs and recommends the investment most likely to maximise output. Perhaps its calculation is technically sound. Yet the advice may still be morally inadequate. Did it consider whether the benefits would be shared? Did it account for the survival of households with the least land? Did it understand that water can be a communal responsibility rather than a commodity allocated to the most efficient user? And who, in the end, was entitled to make the decision: the person typing the prompt, the cooperative’s elected committee, the village as a whole, or generations not yet born?
Such questions expose a fact hidden by the apparent neutrality of computation. An AI system does not merely retrieve information. Whenever it ranks choices, frames a problem or recommends an action, it relies on assumptions about what matters. Those assumptions may concern efficiency, freedom, privacy, ownership, dignity or responsibility. They are ethical commitments even when they appear only as technical defaults.
Most widely used AI systems have been built within institutions based in North America, Europe and parts of Asia. Their design reflects the legal systems, commercial incentives and philosophical habits of those environments. The user is generally imagined as an individual with personal preferences, personal data and personal goals. A good system is expected to protect that person’s rights, increase their options and help them act independently.
These aims can be valuable. But they are not culturally weightless, nor do they exhaust the possibilities of moral life. Across much of sub-Saharan Africa, traditions associated with Hunhu or Ubuntu begin from a different picture: human beings become fully human through their relationships with others. From this perspective, intelligence cannot be measured only by how successfully an individual pursues a chosen end. It must also be judged by whether a person preserves community, recognises mutual dependence and acts in ways that allow others to flourish.
What would artificial intelligence look like if that were its starting point?
Every system contains a picture of the person
Public discussions of responsible AI often revolve around familiar principles: fairness, transparency, accountability, safety, privacy and human control. These standards have helped expose discrimination, secretive decision-making and reckless automation. They provide regulators and engineers with a vocabulary for challenging harmful systems.
Yet ethical frameworks do more than list desirable principles. Beneath every framework lies an answer to a more basic question: what kind of being deserves protection, and what makes that being valuable?
Modern liberal thought often treats the individual as the primary moral unit. A person possesses rights that cannot simply be sacrificed for the group. Consent belongs to the individual; personal information is understood as something over which that individual should exercise control. Political and economic institutions are frequently assessed according to how well they preserve freedom of choice.
This model has supported important struggles against coercion. It helps protect minorities, dissenters and vulnerable people from being absorbed into the will of a majority. Any alternative ethic must take those achievements seriously.
But when individual autonomy becomes the unquestioned foundation of technology, it can make other forms of value difficult to see. A recommendation system asks what you want to watch. A financial application helps you maximise a return. A career assistant identifies the move most likely to advance your ambitions. Even when millions of people use the same platform, the software addresses them one at a time.
Communities then appear in the system as collections of user profiles rather than as moral agents in their own right. Shared customs, collective responsibilities and relationships with land may be reduced to variables describing individual behaviour. Decisions that should involve discussion can quietly become personalised transactions.
This is why claims that AI is universal should be treated with caution. A model can operate in many countries while still carrying a narrow image of the human person. Global availability is not the same as ethical universality.
Language carries more than information
The imbalance is visible at the level of language. Africa contains an extraordinary range of languages, yet only a small proportion receive meaningful support from large language models. Many systems perform well in English and a handful of other dominant languages while producing shallow, inaccurate or unusable results in languages spoken by millions of Africans.
This exclusion has practical consequences. People who cannot use a system in the language they know best may be unable to access educational tools, public information, translation, business assistance or new forms of creative work. Governments and companies may feel pressure to communicate digitally in languages favoured by the technology, reinforcing existing hierarchies.
But the problem is deeper than access. Languages preserve categories of thought, social expectations, humour, memory and ways of describing obligation. A proverb can condense a moral argument that would require pages of abstract explanation. Terms of kinship may encode responsibilities that English words such as relative or family fail to capture. A model that translates the surface meaning while missing the relationships behind it has not truly crossed the cultural distance.
The shortage of digital text in many African languages is often described as a lack of data. That phrase can be misleading. Communities are not empty containers waiting to be filled with machine-readable content. They possess knowledge in speech, performance, ritual, local archives and everyday practice. Turning that knowledge into training material raises questions of permission, ownership, compensation and authority.
Who may record an elder’s account? Can sacred or restricted knowledge be included in a commercial dataset? If a model earns money from a community’s language, should that community share in the benefit? A project that gathers local data without answering these questions may reproduce extraction under the name of inclusion.
Language support, then, is not achieved merely by adding more words. It requires long-term partnerships with speakers, linguists, educators, cultural institutions and local technologists. It also requires the right to decide that certain knowledge should not be collected at all.
Personhood through relationship
Hunhu and Ubuntu are names used in different linguistic and philosophical traditions, and neither term represents the whole of African ethics. Still, they express an influential family of ideas. One widely cited formulation holds that a person becomes a person through other persons. Human identity is not imagined as a finished possession carried into society; it develops through participation in a moral community.
On this view, personhood has an aspirational quality. To be human in the biological sense is not yet to display the fullness of humanity. That fullness is cultivated through generosity, respect, hospitality, responsibility and solidarity. Someone who accumulates power while damaging the community may be successful according to an individualistic measure but deficient according to Hunhu/Ubuntu.
Knowledge also emerges socially. Difficult questions are not always settled by applying a rule in private. They are worked through in dialogue, often with attention to the accumulated experience of elders and the consequences for the group. Deliberation matters because moral understanding is created and tested through relationships.
The community invoked here is larger than the people immediately present. In many accounts, the living remain connected to ancestors, spiritual realities, the natural world and future generations. A decision can therefore be wrong even if it satisfies current preferences. It may violate duties inherited from the past or damage conditions needed by those who will come later.
This differs sharply from the way many AI products represent a decision. A system typically receives a request, identifies the user’s objective and returns an answer. Context is treated as information that improves personalisation. Hunhu/Ubuntu suggests that context is not an accessory to the individual: it helps constitute who the individual is and what choices are morally available.
Return to the farming cooperative. Under a conventional optimisation model, community welfare might be one factor among several, assigned a weight in a calculation. Under a relational model, the cooperative is not merely a constraint on the user’s freedom. It is the setting in which agency becomes meaningful. The decision is good only if the process and outcome sustain relationships of mutual recognition.
Where present-day AI collides with communal life
The conflict appears in at least four areas.
The first is decision-making. Automated systems increasingly influence access to jobs, credit, insurance, healthcare and public services. Even when a machine does not issue the final verdict, its ranking can determine which cases receive attention. Such systems usually evaluate people as separate applicants. They are not designed to pause for a community’s deliberation or to recognise shared responsibility for a hardship.
The second is personalisation. Digital platforms learn what keeps each user engaged and then deliver a tailored stream of information. This can be convenient, but it can also weaken common spaces. Members of the same neighbourhood may receive different political claims, prices or opportunities without knowing that the differences exist. A technology organised around private relevance can erode the shared knowledge required for collective judgment.
The third is data governance. Standard privacy tools ask an individual to accept or reject terms. Yet data about one person often reveals information about others. A genetic record concerns relatives; agricultural data may reveal the practices of an entire village; a photograph can expose people who never agreed to upload it. Relational societies make especially clear what digital systems everywhere tend to obscure: data is rarely individual in its consequences.
The fourth is the definition of progress. AI companies commonly measure success through scale, speed, accuracy, engagement or profit. Hunhu/Ubuntu introduces different questions. Does a system strengthen people’s capacity to solve problems together? Does it distribute benefits fairly? Does it preserve dignity for those who are not profitable customers? Does it honour responsibilities to the environment and the future?
These measures are harder to reduce to a dashboard. That does not make them less real.
Designing for collective agency
An AI informed by Hunhu/Ubuntu would require more than translated interfaces or a ceremonial reference to African values. It would change who participates in design, how goals are chosen and where authority resides.
First, affected communities would help define the problem before engineers selected a solution. Consultation after a system has been built is too late. Farmers, nurses, teachers, informal workers, disability advocates, traditional leaders and local researchers may understand harms and dependencies that are invisible to an outside development team.
Second, some AI tools would need to support group deliberation rather than provide a private answer. A platform used for local planning could present several options, reveal who bears each cost and record areas of disagreement. Instead of predicting the “best” outcome, it might help participants ask better questions. Its purpose would be to enlarge collective agency, not to replace it.
Third, governance would extend beyond individual consent. Communities could establish rules for culturally sensitive data, appoint representatives to oversee its use and retain the ability to withdraw material from a model. Benefits from commercially valuable datasets could be shared through local institutions. These arrangements would not eliminate individual rights; they would add protections for relationships and groups.
Fourth, accountability would be local as well as corporate. A person harmed by an automated decision needs a route to challenge it through an institution they can actually reach. An explanation delivered in technical language from a distant company is insufficient. Schools, clinics, cooperatives and public agencies using AI should retain human responsibility rather than treating the vendor’s output as unquestionable.
Finally, an Ubuntu-oriented system would consider ecological and intergenerational effects. The energy, water and mineral resources consumed by AI infrastructure are not external to ethics. Nor are the working conditions of people who label data or moderate harmful content. A technology cannot plausibly serve communal flourishing while exporting its environmental and human costs to less powerful communities.
The danger of turning Ubuntu into a slogan
There is a risk that this vision becomes romantic. Africa is not a single moral community. The continent contains diverse philosophies, religions, political histories and forms of life. Cities across Africa are shaped by global commerce, migration and strong aspirations for individual freedom. Hunhu/Ubuntu itself is debated, interpreted differently and sometimes invoked by political leaders for self-serving purposes.
Communities can also oppress. Appeals to harmony may silence women, young people, migrants, sexual minorities or dissenters. Elders do not always speak with one voice, and consensus can conceal unequal power. Protecting collective agency must never mean giving unaccountable authorities control over individuals.
The contrast with “Western values” also needs qualification. Western societies contain traditions of solidarity, public service and collective ownership, just as African societies contain competition and individual ambition. The purpose of bringing Hunhu/Ubuntu into AI ethics is not to place two sealed civilisations on opposite sides of a philosophical border.
Its value is diagnostic. It reveals choices that dominant systems present as natural. It asks why the solitary user became the default subject of digital design, why preference satisfaction is treated as freedom, and why communities are consulted only after technologies have begun to reshape them.
A responsible approach would therefore combine relational ethics with firm safeguards for individual dignity. The aim is ethical plurality: institutions capable of listening to different moral traditions without freezing any of them into a stereotype.
A technology that learns to belong
Building such systems will not be easy. The largest AI models depend on concentrations of capital, computing infrastructure and data that few African institutions control. Developers face commercial pressure to produce one scalable product rather than many locally governed systems. Values such as solidarity and consensus are difficult to translate into code, and no technical rule can settle every conflict within a community.
But difficulty is not an excuse for continuing with inherited defaults. AI is still being institutionalised. Laws, procurement practices, datasets and design conventions are being established now. Decisions made during this period will determine whose languages are supported, whose knowledge becomes machine-readable and whose idea of the human is embedded in future systems.
An African AI need not be a separate machine sealed off from the rest of the world. It could instead be a different centre of intellectual and technical authority: models trained through legitimate partnerships, evaluated against locally chosen goals and governed by the people exposed to their consequences. Its lessons would not be relevant only to Africa. Every society faces problems that individualised technology cannot solve, from climate adaptation to public health and social fragmentation.
The deepest contribution of Hunhu/Ubuntu may be its refusal to separate intelligence from relationship. A system is not wise merely because it can generate an answer. Wisdom also concerns how an answer changes the bonds among people, how benefits and burdens are shared, and whether those affected retain the power to deliberate together.
The familiar proverb says that one person cannot encircle an anthill. Neither can one company, one philosophical tradition or one region define an intelligence meant to serve humanity. If AI is to become genuinely global, it must be built through a conversation in which the world is not simply invited to use the technology, but empowered to shape what the technology is for.
