AI Training Is Not AI Literacy, and the Sector Has Been Running the Wrong Programme
UNESCO's new work on civil-service capacity has made the distinction explicit, and it is one every South African organisation should now be able to name.
Sign up for my Substack daily AI newsletter here.
See my AI Training course portfolio for corporate Business Leaders here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
Almost every South African corporate and government department that has rolled out artificial intelligence in the past two years has run some form of AI training. Prompt engineering workshops. Copilot rollouts. Half-day sessions on how to use ChatGPT for productivity. Very few have run AI literacy programmes. The distinction sounds academic. It is not. It is the difference between teaching people how to use a tool and teaching them how to reason about it, and the sector has been mistaking one for the other with increasing consequence.
CONTEXT AND BACKGROUND
The clearest recent statement of the distinction comes from an article published on 3 July 2026 by Yomna Omran, a member of UNESCO’s AI Ethics Experts Without Borders network and a trainer delivering the UNESCO AI Literacy programme in Egypt and Tunisia. Writing from her own experience, she describes what her civil-servant participants tell her at the end of the training. They arrive expecting a course on how to use AI tools. They leave with a set of governance questions they had never previously considered, and a working understanding that AI literacy is a foundation for institutional readiness rather than a skill for individual productivity. The programme itself, developed by the Center for AI and Digital Policy in partnership with UNESCO, is now being delivered across a dozen countries.
The underlying framework is older and worth naming. The 2022 Broadband Commission report, produced by UNESCO and the International Telecommunication Union, set out what AI competencies for civil servants actually consist of, across three domains: digital planning and design, data use and governance, and digital management and execution. The report was explicit that civil servants do not need to become technical experts. They need to understand what an AI system is doing well enough to procure it, oversee it, refuse it where appropriate, and be answerable for the decision either way. The 2025 OECD study Governing with Artificial Intelligence, which examined 200 real-world use cases across 11 core government functions, concluded that the primary constraint on implementation is not budget or technology but civil-service capacity of exactly this kind.
INSIGHT AND ANALYSIS
The confusion between training and literacy is not accidental. It is what happens when a vendor sells adoption and an organisation buys governance. The vendor’s material teaches the tool. The buyer assumes the tool comes with the reasoning. It does not. Literacy, in this context, is an epistemological capacity. It is the discipline of asking three questions of any AI-assisted decision that matters: does this system’s output warrant the confidence I am about to place in it, on what basis, and if it turns out to be wrong who is answerable. Prompt engineering does not teach any of the three. A Copilot rollout does not teach any of the three. A short course in operating a tool does not produce a person who can govern its use, in the same way that a short course on operating a cockpit does not produce a certified pilot. Neither does any productivity course I have seen inside a South African corporate in the past twenty-four months. They teach the tool. They do not teach the reasoning.
I have previously written that in any process where a machine contributes to a decision, a human being has to remain a net contributor to that decision, and accountability has to sit with a named person who can explain and defend it. AI literacy is the capacity that makes such accountability possible. Without it, “human oversight” becomes a procedural checkbox, as Omran puts it plainly, rather than a meaningful accountability mechanism. The consequence is that the organisations pointing at their training records as evidence of governance capability are, in a serious sense, misreading the evidence. Training records show that people can operate the tool. They say nothing about whether anyone in the organisation could recognise when the tool has gone wrong.
Ghana, in March 2026, delivered the UNESCO training programme to civil servants as the first substantive implementation step of its National AI Strategy 2023-2033. South Africa, in April 2026, withdrew a national AI policy whose fabricated references had not been caught in review, and legal commentators here have named “institutional AI literacy” as one of the specific concerns the incident raised. The two events, a fortnight apart on either side of the equator, describe the same gap from opposite ends.
IMPLICATIONS
For South African organisations, the practical implication is that a separate budget line is now needed. Every organisation of any size has an AI training budget, whether formal or informal, and most can name the vendor providing the material. Very few have an AI literacy budget, or a named person accountable for building the capacity the training does not build. The wider implication is one Omran’s work makes explicit and worth naming here. The people who need AI literacy most are not the executives who commissioned the technology. They are the compliance officers, procurement specialists, audit committee secretariat, internal auditors, HR business partners and legal counsel who will actually vet, approve and monitor the deployments. The corporate equivalent of Omran’s insight is that the AI capability conversation has been happening one level too high in the organisation, and the training being provided has been designed for the wrong people.
CLOSING TAKEAWAY
The distinction is not academic. It is the difference between an organisation that can respond honestly when a regulator, an auditor or a court asks how a decision was made, and one that cannot. South African boards, ministers and vice-chancellors should be able to name it plainly this quarter. The next AI adoption programme in your organisation should have two budget lines, not one, and the second should be larger than most people currently assume. Ask who is accountable for the literacy line by name. Ask whether the current programme is teaching the tool or teaching the reasoning. If the answer is that it is teaching the tool, you have training. You do not yet have governance.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments