South Africa's reading challenge puts the educational value of AI under scrutiny - IOL

Angelique Timms, Curriculum Lead at Curro Holdings.

By the end of Grade 3, 15% of South African learners cannot read a single word, a finding that places the country’s foundational learning challenge at the centre of questions about what artificial intelligence can offer schools.

The figure comes from the Department of Basic Education’s Funda Uphumelele National Survey and was highlighted by the 2030 Reading Panel in 2026. The survey assessed 27,838 learners across 710 schools and found that about three in 10 learners in Grades 1 to 3 met their home-language reading benchmarks.

For schools considering artificial intelligence (AI), the question is how its use could help teachers understand where children are struggling, respond to those difficulties and establish whether learning is improving.

Yolandi Farham, Oxford University Press SA, Shirley Eadie, Human Studios and Lefa AI, and Storm Baatjies, MTN, Heather Brogan, Oxford University Press SA.

Reading comprehension, classroom innovation, and responsible AI use formed the focus of Oxford University Press South Africa’s Future Learning Labs, held in Cape Town and Johannesburg in September. The programme brought together teachers, publishing specialists and researchers to explore how digital tools could support teaching and learning.

The discussions connected the possibilities of AI with the practical demands of the classroom. These include identifying gaps in understanding, selecting appropriate support, assessing what learners can do independently, and deciding who is responsible for the material presented to children.

Heather Brogan, Oxford’s digital product manager and the programme’s facilitator, said the starting point should be an understanding of the people expected to use the technology.

“We need to start by asking what a learner is struggling with and what support a teacher needs,” Brogan said during the panel discussion on responsible AI in education. “We can then develop solutions that support learners and teachers responsibly.”

Her emphasis on “what a learner is struggling with” places a specific educational need at the beginning of the process. A school would first identify the difficulty it wants to address, then consider whether a particular tool could help and how its contribution would be assessed.

That distinction matters when a completed assignment offers only a limited view of a child’s understanding. Where AI has helped produce an answer, teachers may need additional opportunities to establish how the learner reached a conclusion and which parts of the work they can explain without assistance.

A discussion after an assignment could provide one such opportunity. Asking learners to describe their reasoning, explain the evidence they used, or apply the same knowledge to a different question could help teachers assess their understanding more closely.

The educational value of that exchange would lie in what it reveals. A learner might be able to repeat an answer but need further support to explain it. Another might understand the central idea but struggle to express it clearly. Follow-up questions could help a teacher distinguish between these difficulties and decide what support to provide.

Reading assessment raises a similar question about how information gathered from learners informs teaching. Angelique Timms, curriculum lead at Curro Holdings, addressed reading comprehension in a session drawing on Curro’s literacy journey.

“A reading assessment should help us decide what to teach next,” Timms said. “When we understand where a child is struggling, we can target the support and check whether it is working. Technology becomes useful when it helps that process.”

The phrase “what to teach next” gives assessment a practical purpose. A result becomes useful when it helps a teacher decide how to respond, while subsequent assessment provides an opportunity to examine whether that response has helped.

Applied to digital learning, this approach would require schools to look beyond whether a resource is available or being used. They would also need to consider what it reveals about learning and whether teachers can use that information to guide their work.

The reading findings make this particularly relevant to the early years of schooling. When many children have yet to meet home-language reading benchmarks, the value of any proposed intervention depends on how well it responds to the skills they are still developing.

The same attention to understanding could inform AI literacy. Vocabulary, comprehension, and subject knowledge may help learners evaluate the answers they receive, recognise where an explanation is incomplete, and formulate questions that take their learning further.

Classroom activities could give children opportunities to practice these skills. Learners might compare an AI-generated answer with another source, identify a claim that lacks supporting evidence, or explain why they find one account more convincing than another.

Such activities would give teachers a basis for assessing the reasoning behind a learner’s conclusion. They could also help make the use of AI part of a wider process of questioning and checking information.

These possibilities bring responsibilities for schools and providers. Decisions about classroom AI involve the suitability of content, the role of teachers, and the arrangements for responding when something goes wrong.

Yolandi Farham, Oxford’s Product Director Africa.

Yolandi Farham, Oxford’s product director for Africa, said responsibility to learners should guide adoption from the outset.

“Every decision to introduce AI carries a responsibility to the learner,” Farham said. “We need to be clear about who checks the content, how teachers guide its use, and how concerns are addressed. Those responsibilities should be understood from the beginning.”

Putting those expectations into practice could involve agreement among school leaders, teachers, and families about what constitutes acceptable assistance. Learners would need to understand when AI use should be disclosed and what information may be entered into a tool.

Expectations could also distinguish between different classroom tasks. The assistance appropriate for exploring an idea may differ from what a teacher permits when assessing independent understanding. Explaining the purpose of an activity could help learners understand the boundaries that apply.

Teachers would need support to exercise the oversight expected of them. Training and time to review generated material could help them assess its accuracy, suitability for the learners concerned, and relevance to the curriculum.

Simple arrangements for raising concerns would form another part of that work. Questions about content or use would need a defined review route, with responsibilities understood by those involved.

The circumstances of South African classrooms were central to the contribution from Shirley Eadie of Human Studios and Lefa AI. She emphasised the importance of involving those affected by decisions about technology.

“South Africa needs an approach informed by its own classrooms,” Eadie said. “Teachers, learners and parents should help identify the opportunities and the risks. Language, access and the realities of teaching must be part of that conversation.”

Her call for an approach “informed by its own classrooms” brings attention to the conditions under which a digital resource would be used. A school considering a tool would need to examine whether learners could access it, whether its language and content suited their needs, and whether teachers had the support to use it effectively.

Yolandi Farham, Oxford University Press SA, Shirley Eadie, Human Studios and Lefa AI, and Storm Baatjies, MTN.

For a child approaching the end of Grade 3, that contribution must ultimately be visible in learning itself, through a growing ability to read, understand, explain and question.