By Professor Darelle Van Greunen
But without these supports — particularly access, literacy and teacher empowerment — AI will almost certainly widen them. The choice is ours.
From access to opportunity
In the early days of the internet, the digital divide was measured by who had devices, connectivity and basic digital skills. The AI era has redrawn that boundary. Today, the divide is defined not by access to technology itself, but by access to opportunity through AI — who can meaningfully use intelligent tools to learn, create and thrive.
Equity in AI-driven education now depends on three interconnected dimensions:
-
Access — the availability of AI-enabled platforms, affordable data and reliable electricity
-
Use — the ability to engage with AI critically and ethically
-
Outcomes — whether learners actually achieve improved performance, employability and confidence.
Without deliberate intervention, South Africa risks replicating the inequalities of the past in a new, digital form.
The risks: cost, capacity and context
AI’s rapid advance has introduced both promise and peril. Affluent schools and universities can afford premium tools, faster internet and specialised AI training. Learners in under-resourced schools often share mobile phones, face high data costs, or study through load-shedding. Many are “digitally social” but not “digitally skilled” — comfortable on social media yet unprepared to use AI responsibly or effectively for study or work.
Meanwhile, teachers — many of whom never received digital training — are expected to navigate AI-powered classrooms with little support. Uneven digital literacy, combined with limited infrastructure, deepens disparities. Language and cultural bias compound the problem: most mainstream AI systems are trained on non-African data, leaving local learners underserved and sometimes misrepresented.
When AI levels the playing field—and when it doesn’t
The contrast between well-resourced and low-resourced contexts is already visible. Premium AI tutors and subscription-based models give wealthy students tailored feedback and exam coaching, while others are left behind on free, generic tools that often fail to align with the South African curriculum.
Even small mismatches can matter. When AI tools offer explanations drawn from international syllabi, they risk confusing learners about local content or assessment standards. To prevent this, technology must be context-aware — rooted in African curricula, languages and lived realities.
There are, however, promising examples of AI being used to reduce inequality. Zero-rated learning portals and WhatsApp-based tutoring services are extending access to rural and township learners. Accessibility tools such as speech-to-text, captions and translation are enabling more inclusive learning for students with disabilities and multilingual backgrounds.
These examples demonstrate that with thoughtful design, AI can open doors rather than close them.
Designing an equity-first AI ecosystem
The most urgent need is for a national, equity-driven AI-in-Education framework that aligns with South Africa’s privacy and protection laws (POPIA). This framework should prioritise model transparency, bias testing, and cultural inclusivity.
Equitable licensing must follow: national negotiations with technology providers could secure low-cost or zero-rated access to approved AI platforms for all public schools and universities. Similarly, device-plus-data bundles, public AI labs and reliable campus Wi-Fi would help bridge the physical access gap.
Another essential intervention is teacher enablement. AI literacy must become part of continuous professional development. Tiered micro-credentials — from basic awareness to advanced classroom integration — can ensure that teachers learn at their own pace, supported by provincial and university networks. For students, AI clubs, WhatsApp micro-lessons and model lesson banks can promote hands-on, peer-driven learning.
Assessment reform must accompany these efforts. Education systems need to recognise process, critical thinking and creativity rather than simple content recall. AI can assist with feedback and differentiation, but validation must remain human-centred and integrity-focused.
Bias, privacy and the problem of design
The fairness of AI systems hinges on how they are built. Models trained largely on Western data often fail to recognise African names, accent, and contexts. This erodes trust and can unintentionally perpetuate stereotypes. Mandatory bias testing on South African datasets, along with inclusion of African languages and examples, is critical.
Data privacy poses another serious concern. Without strict protection, minors and vulnerable learners are exposed to tracking and exploitation. POPIA-compliant safeguards — data minimisation, local hosting where feasible and robust consent protocols — must be non-negotiable.
Finally, design for accessibility must go beyond aesthetics. Tools that assume constant high-speed connectivity exclude most rural and peri-urban learners. Developers should prioritise offline, low-data and mobile-first interfaces, with vernacular options that reflect South Africa’s linguistic diversity.
AI literacy for all
For AI to be a genuine equaliser, all stakeholders — students, teachers, administrators and parents — need a baseline understanding of what responsible use looks like. Beyond technical competence, AI literacy includes fact-checking, bias detection, ethical use and privacy awareness.
In practice, this means creating short, modular learning opportunities for teachers, embedding AI content into teacher education programmes, and promoting lifelong learning. For students, practical exposure is key: using AI to explore topics, design projects or analyse data under guided supervision builds confidence and critical awareness simultaneously.
The economics of access
The cost barrier cannot be ignored. Subscription-based AI tools threaten to entrench privilege by limiting full functionality to those who can pay. The solution lies in public-interest, locally hosted AI models, open-source innovation, and equitable licensing negotiated at national level.
Universities can lead by hosting shared compute infrastructure, training open multilingual models, and offering campus access to premium tools for disadvantaged students. Partnerships with telcos and NGOs can enable zero-rated educational access, ensuring that no student’s learning is capped by their data balance.
A shared future — or a divided one
If current inequalities persist, AI will deepen them. The risk is clear: elite schools and universities will accelerate ahead, while rural and township learners remain disconnected. Assessment fraud will rise, trust in qualifications may weaken, and foreign pedagogies could displace local contexts.
Yet, with deliberate action, AI could help transform South African education into one of the most inclusive systems in the world. Imagine every learner having access to a low-data, multilingual AI tutor aligned with the national curriculum? Every teacher equipped with AI-driven tools to plan lessons, differentiate instruction and lighten administrative loads?
AI will not determine the future of education — our choices will. Equity, access and local relevance must remain the compass points. If these guide the journey, AI can help shape a more just, inclusive and future-ready education system — one where no student, teacher, or community is left behind.
Professor Darelle van Greunen is Distinguished Professor of Information Technology and Director of the Centre for Community Technologies, Nelson Mandela University, South Africa