The Just AI Conference began with a robust selection of parallel discussions in which multidisciplinary researchers presented papers across four themes: democratic futures in the algorithmic age; AI extraction to AI restoration; building Just AI in DPI; and relational ethics and contextual knowledges in AI design. The insights and case studies discussed left us with five key reflections, each of which confirmed that justice is more than a normative principle – it’s a technical, social, and political imperative.
When policy reaffirms systemic power imbalances, regulation only constrains us further
Loise Ochanda opened the day with a live policy design problem: Kenya’s draft AI Bill. The document has generated significant discussion, not because regulation is unwelcome, but because of who controls the regulatory architecture – and in whose interest. She raised concerns around how the Bill positions AI as a significant developmental pathway, but the energy- and water-hungry data centres it would require remain limited. Dr Angella Ndaka, Keynote Speaker, later reiterated these concerns, further highlighting how the Bill risks overregulating small African startups while global tech companies, which hold disproportionate influence over regulatory design itself, continue to operate with limited accountability.
Without justice-oriented technical design, past bias will perpetuate in future data
AI-powered credit scoring draws on historical loan access data. But that data reflects decades of exclusion from formal financial services – particularly for rural women. The algorithm scales these existing structural inequalities with added opacity and authority. Ambassador Dr Lavina Ramkissoon underscored this point when she said: “Code isn’t neutral, and it shouldn’t be.” Just AI Fellow, Emmanuel Kpakpo Brown, pushed this further: “When authority shifts from courts to code, justice must be embedded within systems”. For example, with terms, conditions and warranties routinely designed to be too long for users to meaningfully engage with, we need systems that prompt users in local languages, including voice alerts, so that consent can be informed rather than assumed. The emergent policy question is not only how we mandate contextually grounded bias audits, but also how we build forensic-ready systems where justice is a technical design requirement, and where organisations can be held accountable for any shortcomings.
Language extractivism is both sociotechnical violence and neocolonial capture
“Who owns the right to your mother tongue? Are there models that run on your mother tongue? If those models exist, how do they perform?”
Dr Melissa Omino, Leonida Mutuku, Mark Irura and Loise Mboo provoked the audience with this question, pointing out that communities are actively contributing language data – often through micro-task platforms, with very little payment – to train models they will never own, and maybe never use. Looking at a proposed Kenya Language Corpus framework, developed through a qualitative, participatory methodology centred on multi-stakeholder engagement, they explored a different model: tiered benefit-sharing, sovereign public funding, and language communities recognised as rights holders rather than data sources. What emerged was a governance design challenge: how do you build a language data repository with a decades-long lifespan, protected from both government switches and private capture, that keeps value flowing back to contributing communities? This is a vital question for a continent with over 2000 languages. As Dr Tigist Shewarega Hussen said: “Language AI emerges not merely as a technical pursuit but as an ongoing sociotechnical act of decolonisation.” When we reclaim our mother tongue, we not only retain our languages but also preserve our cultures, embodied histories, traditions, indigenous knowledge, and philosophies.
AI Strategies and policies can provide development-oriented pathways, but we need the resources to operationalise them
Verengai Mabika showed us how Zimbabwe’s National AI Strategy, aligned to the country’s Vision 2030, offers an instructive example of a development-oriented approach to AI governance – positioning AI as a driver of knowledge-based economic growth, in the same way that South Africa’s National AI Policy Framework frames AI as a tool to position the country as a global leader. But as Sally Ncube, Keynote Respondent, cautioned, optimistic strategy documents face a consistent obstacle: the ‘subject to availability of funding’ clause. We know what we need to do, but without the resourcing architecture we need, implementation and enforcement become unlikely. Further, we need the right international development cooperation models that support African agency rather than simply reinforcing donor dependency.
AI harms are cumulative and largely unaccounted for
If an AI system misdiagnoses a patient, and that misdiagnosis only becomes apparent twenty years later, what does redress look like? What are our approaches to post-harm accountability?
Existing AI governance frameworks are frequently built to anticipate AI harms in advance, using ex-ante regressive or preventive approaches to address risks before AI systems are even deployed. But in reality, these harms are frequently slow, cumulative, and uncertain. Consequently, we need legal or institutional infrastructure to acknowledge these harms, repair the relationship between the affected person and the system that failed them, and reform the institutions responsible after harms have occurred. Thinking practically, Jonas Kgomo proposed a Redress Clause that comes into play once harms are observable in practice rather than merely speculative. In African contexts defined by structural inequality and limited institutional capacity, this becomes particularly important, given our need for adaptive mechanisms to acknowledge harm rather than rely on paper-based risk assumptions and management.
Just AI must reflect the diversity of the continent and its people
The thread running through all of these cases is the same: Just AI governance in Africa cannot be a checklist of principles appended to existing ICT policy. It requires confronting structural inequality directly – in system design, regulatory architecture, procurement, and resourcing models – to ensure that Africans remain in control of the technologies they use, rather than mere resources and rule-takers for the technologies deployed to us.
