SA schools are failing — but new research from Nigeria offers hope

News24

22-06-2025

A new working paper from Nigeria offers a glimpse of what an artificial intelligence-augmented education might look like. In Benin City, researchers ran an experiment in nine ordinary public schools. The intervention was modest, but the results were anything but, says Johan Fourie.
Anyone who occasionally opens a South African newspaper will know the story: most of our public schools are failing. But for those who have grown numb to these stories, a few numbers should still jolt you awake. Four out of five Grade 4 learners cannot read for meaning in any language. One in four learners will repeat Grade 10. One in five teachers is absent on Mondays and Fridays, and one in three are absent at the end of the month.
These are not new problems. I distinctly remember sitting in a development economics class in the early 2000s, wrestling with the challenge of South Africa's poor education outcomes. Back then, the prevailing hope was that these were teething issues: a temporary hangover from the transition to democracy, soon to be fixed by a public school system finding its feet. I was optimistic that steady investment and reform would deliver the turnaround we needed. But two decades of painfully slow progress have forced me to reconsider. I now find myself firmly in the voucher camp. The idea is straightforward: instead of allocating government funding directly to schools, each child would receive a voucher – say, R20 000 per year – which they could use to enrol at the school of their choice, public or private. The logic is that real choice, backed by state support, would foster healthy competition between schools, driving up quality far more quickly than our current approach. If anything is going to shake us out of this rut, it is giving families the power to choose.
But ideas like that rarely get off the ground, not because they lack merit, but because there are too many vested interests determined to block any radical change. Powerful groups stand to lose if old certainties are upended, and so real reform is strangled before it even begins. Any solution to our education crisis must, somehow, find a way to keep almost everyone on board – at least at the start.
So let me offer one solution. A new working paper from Nigeria offers a glimpse of what an artificial intelligence-augmented education might look like. In Benin City, researchers ran an experiment in nine ordinary public schools: invite students in for after-school English lessons, but with a twist. Instead of a chalkboard, each child sat at a computer, learning English with the help of ChatGPT-4. The intervention was modest – just 12 sessions over six weeks – but the results were anything but. By the end, students who attended the AI-supported classes outperformed their peers by more than 0.3 standard deviations on the final assessment. That might sound modest, but it equates to nearly two years' worth of typical learning gains.
Crucially, these were not just superficial improvements in digital skills or answering AI-generated trivia. The most significant gains were in English itself. The researchers found that students who participated in the AI tutoring also scored much higher on their standard end-of-year English exams – the same exams taken by all students, covering the entire year's curriculum. As the authors put it, the impact of the intervention extended well beyond the brief six weeks of the programme. In simple terms: a short burst of AI-assisted tutoring delivered learning improvements that our traditional schooling system has failed to achieve in years.
There are subtleties worth noting. The programme benefited all students, regardless of their starting point, but the strongest effects were seen among those who had performed better before, and among girls, who started with lower scores but made up ground. The more sessions a child attended, the bigger the gains: the researchers estimate that a full year of AI-augmented tutoring could boost achievement by up to two standard deviations, a remarkably large effect.
And the real surprise: it was almost absurdly cheap to run. When the researchers did the maths, they discovered the programme yielded the equivalent of more than three years of schooling for every $100 spent, an efficiency that leaves traditional interventions in the dust. For a country, or even a continent, grappling with limited resources and a rapidly urbanising population demanding education, results like these shift the boundaries of what's possible.
If one study is not enough to convince you, then let me add a second, meta-analysis study, published just last month in Nature Humanities & Social Sciences Communications. The authors summarise the results of 44 research papers reporting the impact of ChatGPT on learner improvement. Their conclusion? The 'calculated effect size indicates a large positive impact'. In short: yes, ChatGPT boosts education outcomes significantly.
So, if I were to advise the Minister of Basic Education, here's what I would recommend we implement tomorrow: make free AI tutors available to every child in a public school. The early evidence is persuasive enough that, for once, this is a bet worth taking.
But I suspect it won't be nearly so straightforward. The Minister will inevitably run into the same old obstacle: vociferous, well-organised teacher unions with plenty of political clout. Sensing that AI tutors might expose just how poorly their members have been performing, the unions will do everything possible to block implementation. They'll point out, quite rightly, that Large Language Models can hallucinate, but quietly sidestep the mounting evidence that many teachers in under-resourced schools are just as likely to get things wrong. They'll insist that education is about more than rote learning; it's about instilling values, fostering citizenship, and shaping the next generation. In theory, this is correct, but in practice, it rings hollow when so many teachers struggle to attend a full week of lessons. When pressed, unions will argue, as they have before, that wider factors in the classroom prevent teacher success, yet they remain reluctant to experiment with alternative approaches that might address these very issues. The result is all too familiar: rhetoric about the sanctity of education, and resistance to anything that might unsettle the status quo.
So, here's a second proposal for the Minister: bring the teacher unions on board by paying them to help implement AI tutors in a random selection of schools across the country. Guarantee the salaries of all teachers for the next five years as part of this national pilot. No one loses out, and everyone has the chance to see what works. Even better, equip the participating teachers with their own AI tutors, allowing them to develop new skills alongside their students. At the end of five years, if education outcomes show no improvement, the programme can be cancelled. But if the results match those from Nigeria, with a clear and substantial boost, then it should become the new standard everywhere.
AI tutors could also help address another looming crisis: as the teaching workforce ages and fewer young people choose the profession, South Africa is heading for a serious teacher shortage. AI will not displace the need for a human in the classroom, but it can ease the load, allowing larger classes without sacrificing personalised support. This is already happening: South African company Mindjoy is building AI tutors that deliver adaptive content, real-time feedback, and 24/7 personalised learning, while automating routine tasks to free up teachers for deeper engagement. Instead of the old classroom model, imagine dynamic learning spaces where students of all ages learn at their own pace, guided by teachers and AI working together. The future of education could be both more flexible and more effective – if we have the imagination and political courage to build it.
If we accept that four out of five children still leave primary school unable to read for meaning, that thousands of teachers are absent on any given day, and that a quarter of learners are repeating grades, we are simply condemning another generation to disappointment. Clinging to the status quo is a decision, one that chooses failure by default.
But if we treat technology as the ally it can be, and summon just a fraction of the ambition our children deserve, there is no reason why South Africa's schools must remain at the bottom of the rankings. The numbers can change, if we do.