TAYLA PINTO | Why SA needs AI-literate MPs

When I train MPs on AI I open with a simple question: do any of you engage with AI? Most say, “No”.

They are wrong! The reason they are wrong goes to the heart of the problem: there is a fundamental misunderstanding of what AI is and how deeply it is already integrated into our lives.

Many of them had used AI before breakfast that morning. It curated the videos YouTube suggested to them, filtered the spam out of their inboxes, and ranked the results when they googled a constituent’s query.

AI is not something they have declined to adopt. It is something they are already immersed in, invisibly and constantly, without consent forms or announcements. What they mean when they say “No” is that they have never opened ChatGPT.

That confusion, between AI as a chatbot you visit and AI as an infrastructure you live inside, is the most important thing I have learnt from engaging and facilitating discussions with legislators who will govern this technology.

It matters because you cannot oversee what you cannot see. A parliamentarian who believes AI is some remote, even dystopian Silicon Valley curiosity that other people use is most likely not to think to ask whether it is embedded in departmental procurement processes, policing infrastructure or a system that decides who receives social welfare. The greatest governance risk this technology poses is not its power, but its quietness.

South Africa has just been given a costly demonstration of what that quietness can do. In April communications & digital technologies minister Solly Malatsi withdrew the draft National AI Policy, a cabinet-approved document already published for public comment, after journalists discovered that at least six of its 67 citations were fabricated.

These fabrications were AI-generated references to articles that were never written, attributed to real academic journals and in some cases to nonexisting journals. The document meant to establish how South Africa governs AI was itself undone by unverified AI.

The minister called it an “unacceptable lapse” and it was certainly that. Yet it was also something more instructive: an important lesson on the fact that AI does not fail the way other technology fails. It does not crash or display an error message. It produces fluent, authoritative, professional-looking text that happens to be false. No-one in the review chain caught it. Why? No-one thought to check it due to this fluency.

To parliament’s credit, the chair of its communications committee was among the first to publicly call for the draft’s withdrawal. That is oversight at work. However, it kicked in only after journalists had done the checking and after cabinet approval and publication. This should warrant a question we have been too modest to ask. How many of the people conducting that oversight could explain what an AI hallucination is, why it happens, under what conditions and where else in the government the same failure could be sitting?

This is why I have come to see AI literacy not as professional development for parliamentarians but as part of the job itself. Oversight is a constitutional function. An MP who cannot ask what data a system was trained on, whether its outputs can be explained and who is accountable when errors are picked up cannot meaningfully scrutinise a department that deploys AI.

They cannot interrogate a procurement contract dressed up in claims of efficiency and innovation. They cannot protect a constituent whose grant application, job screening or police profile was shaped by a system no-one in the room understands. AI illiteracy in parliament is not a knowledge gap; it is an oversight gap.

The aim is not to turn public representatives into technologists any more than budget oversight requires them to be chartered accountants. It is to empower them with enough understanding to ask the following questions that make quiet systems visible:

• What problem is this meant to solve?
• What information does it rely on?
• Can its decisions be explained and challenged?
• Who benefits, who may be harmed, and who answers when it gets something wrong?

In my training sessions this shift happens quickly. Once people recognise the AI already within their reach, they start recognising it in the state and they start asking better questions in hours. The capacity is there. What has been missing is AI literacy; the exposure.

The window for building that capacity is now, not after the next failure. Systems being procured now will be embedded in public administration for years. Once an automated process is built into a department it becomes expensive to remove and easy to stop questioning.

We should not wait for a discriminatory automated decision or a failed multimillion-rand system before asking what responsible adoption looks like, especially when the AI policy episode has already shown us, cheaply by comparison, how easily unexamined AI slips through.

So, here is a concrete proposal. Parliament should make AI literacy a standard part of member induction and capacitating committees. This should be treated in the same manner it treats budget literacy. Not as an optional extra or a concern for specialists, but as foundational to the work of holding a modern state to account.

New members go through an induction programme, sitting members attend capacity-building workshops on the budget process and oversight, and a dedicated parliamentary budget office supports fiscal scrutiny. Nothing equivalent exists for algorithmic scrutiny. Parliament does not need to build anything new. It needs to point its existing training infrastructure at a new subject.

The timing is not abstract either. A panel of researchers, lawyers and governance specialists has already been deployed to rebuild the withdrawn AI policy and portfolio committees across health, policing, education and social development will increasingly review systems with algorithms inside them. The members tasked with this work should be equipped before those documents reach them, not embarrassed after.

The future of AI in South Africa will not be decided by technology companies alone. It will be shaped by the people who write our laws, oversee our institutions and allocate public resources. They do not need to have all the answers. But when I ask a room of them whether they use AI, the answer needs to stop being no, because the systems are already here, already deciding things, and already waiting to be asked.

• Pinto is an attorney turned legal engineer who develops and delivers AI literacy materials and training.

Business Day