🤝 Ubuntu & Relational Ethics

Core Principles

"Umuntu ngumuntu ngabantu" — a person is a person through other people. Ubuntu holds that:

• Personhood is relational: You become fully human through your relationships with others, not in isolation
• Interdependence is fundamental: Individual flourishing depends on community flourishing — and vice versa
• Communal responsibility: Ethical obligations extend to the community, not just to individuals directly affected by an action
• Dignity through belonging: Every person has inherent worth, which is realised through meaningful participation in community life
• Restorative orientation: When harm occurs, the goal is to restore relationships and reintegrate, not merely to punish

Ubuntu vs. Individualist Ethics

The contrast with dominant Western ethical traditions is not about one being "better" — it is about fundamentally different starting points that reveal different things:

• Starting point: Individualist ethics begins from the autonomous individual and asks about rights and obligations. Ubuntu begins from relationships and asks how actions affect the web of social connections.
• Unit of analysis: Western frameworks often assess harm to individuals. Ubuntu assesses harm to relationships and communities.
• Concept of justice: Rights-based frameworks focus on fair procedures. Ubuntu asks whether the social fabric has been strengthened or weakened.
• Accountability: Individualist ethics assigns responsibility to specific actors. Ubuntu understands responsibility as distributed and shared.

Neither framework sees everything. Each has blind spots that the other illuminates.

The Limits of Rights-Based Frameworks

Mhlambi identifies several things that individual rights frameworks struggle to address in the context of AI:

• Collective harm: When AI systematically underrepresents a community in training data, the harm is to the community as a whole — not easily reducible to individual rights violations
• Erosion of social trust: AI-generated misinformation damages the shared epistemic commons, but "social trust" is not an individual right that can be protected through conventional legal frameworks
• Cultural erasure: AI systems trained primarily on English-language data from the Global North can homogenise knowledge and expression — a harm to cultural diversity that individual consent frameworks do not capture
• Power asymmetries: When communities in the Global South are subjects of AI systems designed in the Global North, the power relationship is structural, not individual

What Relational Ethics Adds

An ubuntu-grounded approach to AI ethics asks different — and additional — questions:

• Relational impact: How does this technology affect relationships between people? Between communities? Between knowledge traditions?
• Community bonds: Does this AI use strengthen or weaken the communal bonds through which people flourish?
• Inclusivity of consideration: Who is excluded from the community of those whose wellbeing matters? Whose voices are absent from the design and governance process?
• Power dynamics: How does this technology redistribute power? Does it concentrate control or distribute agency?
• Restorative potential: When AI causes harm, how can relationships be repaired — not just individual compensation paid?

The Argument

Algorithmic injustice cannot be fixed by debiasing individual systems. The conventional approach — identify bias in a dataset, remove it, retrain — treats injustice as a technical bug to be patched. Birhane argues this fundamentally misdiagnoses the problem.

Algorithms do not merely contain biases. They encode and amplify existing power relationships. A facial recognition system that performs poorly on dark-skinned faces is not just "biased" — it reflects and reinforces a world in which certain faces are valued more than others. The injustice is relational and structural, not just statistical.

A relational ethics approach asks not just "is this algorithm biased?" but "what relationships of power does this algorithm create, sustain, or disrupt?"

Implications for Researchers

When using AI in your own research, a relational lens prompts questions that technical frameworks do not:

• Whose knowledge? Whose knowledge was used to train this model? Whose perspectives are represented — and whose are systematically excluded?
• Who benefits? Who benefits from the outputs of this AI-assisted research? Are the communities whose data contributed to the training also benefiting from the results?
• What is reproduced? What existing power relationships does this AI tool reproduce or amplify in my research context?
• Whose categories? When AI applies categories to my data (thematic codes, classifications, groupings), whose conceptual framework is it applying? Is that framework appropriate for my research context?

Why "Just" AI — Not Just "Responsible" AI

The RIA Framework (Chetty & Sey, 2025) makes a precise distinction. Widely circulated global frameworks — including the EU Ethics Guidelines for Trustworthy AI, the OECD AI Principles, and UNESCO's Recommendation on the Ethics of AI — advocate for qualities like accountability, fairness, transparency, and trustworthiness. These are necessary conditions, but the framework argues they are insufficient to guarantee just outcomes.

The example is sharp: an AI tool made available across all socioeconomic backgrounds may still produce unjust outcomes if a skills deficit prevents users in under-resourced contexts from leveraging it effectively. The tool is "fair" by conventional metrics, but the outcome is not just.

Justice, in the RIA framework, is explicitly restorative and redistributive — not merely procedural.

Why an African Framework Matters Globally

Current AI ethics discourse is dominated by institutions in the Global North. This produces governance frameworks that encode specific epistemological assumptions, legal traditions, and economic interests that do not transfer straightforwardly to African contexts — or, arguably, to many other contexts beyond the societies in which they were developed.

• Unique developmental challenges: Infrastructure deficits, diverse democratic systems, colonial legal legacies
• Risk of consumer status: Without deliberate African-led governance, the continent risks being positioned as a consumer rather than a co-creator of AI technology
• Epistemic justice: African knowledge systems, ethical frameworks, and lived experiences must be foundational to governance — not afterthoughts

1. Human Rights Primacy

Are AI systems developed and deployed in compliance with human rights frameworks? Are human rights impact assessments mandated at each stage of the AI lifecycle?

2. All of Humanity

Is the AI value chain globally inclusive? Do mechanisms exist to ensure that marginalised countries' voices and needs are incorporated into AI governance?

3. Structural Inequality

Do AI systems account for and challenge structural inequalities, rather than encoding and amplifying the historical biases embedded in training data?

4. Equity & Inclusion by Design

Are equity and inclusion built into every stage of the AI lifecycle — from problem formulation through deployment and monitoring — rather than retrofitted after the fact?

5. Ethics of Care

Are AI systems grounded in context and relationality? Do they prioritise human relationships and wellbeing — especially for less-powerful groups?

6. Democratic Governance

Is the AI value chain subject to democratic oversight and control? Are there mechanisms for meaningful public participation in AI governance?

7. Economic Justice

Are AI's benefits shared equitably? Are those negatively impacted compensated or supported? Does governance achieve restorative, redistributive justice?

8. Data Justice

Do AI systems respect personal and community data? Are both individual and collective data rights recognised? Is data used to empower rather than extract?

9. Sustainability

Does the AI value chain consider environmental and social impacts? Are assessments of environmental costs required — and acted upon?

Power Asymmetries

AI systems are overwhelmingly designed in the Global North and deployed globally. This creates structural asymmetries that conventional "fairness" frameworks do not adequately address:

• Training data: Massively underrepresents Global South languages, knowledge systems, and contexts
• Benefit distribution: Economic benefits of AI accrue disproportionately to technology-producing regions
• Governance voice: African and Global South countries have limited influence in global AI governance forums
• Infrastructure dependency: Reliance on cloud computing infrastructure owned and operated by foreign companies

Local Knowledge and Data Sovereignty

Critical questions about knowledge and power:

• Data ownership: Who owns data about African communities? Who profits from it?
• Community consent: Should AI trained on local knowledge require community consent — not just individual consent?
• Benefit sharing: If AI systems generate value from local data, should communities share in the economic benefits?
• Epistemic justice: The RIA framework's insistence that African knowledge systems must be foundational — not afterthoughts — to AI governance is both a theoretical and a practical demand

From Local Realities to Global Forums

The RIA Just AI Framework is explicitly designed to bridge this gap — generating evidence robust enough to inform national policy while also carrying weight in international governance dialogues. The aim is to shift Africa's position from a relatively weak voice in global conversations to an active and dynamic site of knowledge production, resistance, and governance.

As researchers at a South African university, you have a particular vantage point on these questions — and a particular responsibility to engage with them.