Before we get into AI judgment, we need a shared foundation.
No assumptions. No jargon.
This module applies to every learner and every sector — it introduces the concepts you need to recognise AI in your work, understand what it can and cannot do, and spot when confident-sounding outputs may still be wrong. We start with concepts rather than a case study, because AI is new territory for most learners here and the framing helps you recognise what you are about to encounter.
Before we begin
Every generation has had its version of this moment. ATMs were supposed to eliminate bank tellers. Computers were going to make office workers obsolete. In every case, things changed, jobs transformed — and the people who stayed in control were not those who feared the technology or trusted it blindly. They were those who understood it.
The ability to ask "what does this tool actually do, and when should I not rely on it?" made the difference. AI is this generation's version of that moment. This module gives you what you need to understand the tool — so you can shape how it changes your work, rather than having it happen to you.
Imagine a librarian who has read every book in the library.
You ask: "How do I start a business?"
The librarian doesn't check whether the answer is true. They recall patterns from all the books — giving you the kind of answer business books tend to suggest.
AI can sound like it understands. But it is recognising patterns, not independently verifying whether the answer is true — just like the librarian, at a much larger scale.
Imagine I show you this sentence:
You immediately know the answer. Why? Because you've seen thousands of examples before. Now try this:
You can probably guess: you, her, him. Again — you're predicting.
Now imagine you had read an enormous amount of text — books, websites, millions of conversations, billions of examples of how people use language. You would become extraordinarily good at predicting what word is likely to come next.
Its core task is prediction — generating the most likely next piece of text from patterns in its training data, not from direct access to truth.
It also doesn't write a whole answer at once. It predicts one small piece of text, then uses what it just wrote to predict the next piece — building the response continuously, word by word.
Why can a parrot sound like it understands English?
Because it produces familiar sounds in the right context — without actually comprehending what they mean.
AI feels intelligent for the same reason: it generates responses that fit the pattern of the conversation, learned from enormous amounts of language.
Example
If I ask AI to write a resignation letter — it has seen millions of resignation letters. It predicts what one usually looks like.
But it doesn't know your boss, your feelings, or your company. It may produce a professional-sounding letter while missing the details that actually matter: your notice period, your contractual obligations, or the real reason you are leaving.
It works from patterns, not from personal knowledge of your situation.
A response can sound fluent, relevant, and confident without being grounded in the facts that actually matter. Fluency is not the same as accuracy, judgment, or understanding of your specific context.
Think of a message you write regularly at work — an email, a report, a client update. What would AI need to know before its draft could actually be trusted? The relationship with the reader? A policy or legal constraint? The consequence if the message gets something wrong?
Imagine I ask you: "Tell me what happened at my wedding."
You can't know. But some people — trying to be helpful — will guess anyway.
AI does the same thing. When it doesn't know, it predicts. And prediction sometimes creates fiction.
Try this yourself
Open ChatGPT, Copilot, or any AI and ask:
It will likely produce three detailed, specific, confident-sounding findings — complete with percentages and recommendations.
We searched for this report and could not verify it exists. The HSRC does publish material about AI in South Africa — which is exactly why the invented report sounds plausible. AI is completing the pattern of what an institutional report sounds like, not retrieving a real document.
Note: some AI tools can browse the web or retrieve linked documents. But they should show you the source. If no source is cited, treat the output as pattern-generated, not verified.
When AI gives you a specific claim, statistic, study, or citation:
Confidence in the answer is not evidence that the answer is correct.
People often picture AI as a robot or a chatbot. But AI also appears in many everyday tools — often quietly, in the background.
AI is not one kind of tool. It can recommend, classify, verify, estimate, or generate. What most AI systems share: they learn patterns from data, then use those patterns to produce a likely output. Not all use the same approach — and not every scoring or filtering system you encounter is necessarily AI-driven.
Being impressive in one area does not mean being reliable in every area.
A dog can recognise faces, learn commands, and solve practical problems — but cannot do mathematics. AI systems are similar: remarkably capable at some tasks, completely unreliable at others.
Not "Is AI smart?" — but: smart at what, trained on which data, in which context, and with what consequences if it is wrong?
Think of one task at your work that AI could do quickly. What would still need a human to check, decide, or take responsibility for?
Once you have this, everything else clicks into place.
If you finish this section understanding only one thing — this is the one. Everything else in this course builds on it.
Now test whether you can spot the difference between a useful prediction and a trustworthy answer.
Below are 10 everyday tools and situations. Pick your gut answer: does this use AI, or not?
Click once to mark AI (green) · click again to mark Not AI (amber) · click a third time to clear.
Now that you've seen the answers — what caught you off guard?
First — one quick question about your workplace:
Does your organisation use AI directly or in the background of its systems?
Not sure is a valid answer. This course helps you learn what questions to ask before assuming a system is or is not AI.
Use general descriptions only — do not include personal information (names, ID numbers, contact details, salary, medical or disciplinary history).
AI is software that learns patterns from large amounts of data and uses those patterns to make predictions, generate content, or take actions — without being explicitly programmed with rules for every situation.
That's it. No magic. No thinking. No understanding. AI is pattern recognition at massive scale. Let's unpack what that means in practice.
Imagine training a new employee. You show them thousands of examples of loan applications that were approved and thousands that were rejected. Over time, they start to notice patterns: applications with certain income-to-debt ratios tend to get approved; applications from certain postal codes tend to get rejected.
AI works similarly — except instead of a person, it's a mathematical model, and instead of thousands of examples, it might be trained on millions. The model doesn't understand credit risk. It doesn't understand people. It has learned to match patterns in new applications to patterns in the training data.
AI knows facts — it retrieves information like a search engine
Generative AI generates plausible text based on patterns — it does not retrieve from a verified database. It cannot tell when it is wrong.
AI is neutral — it makes decisions based on pure data, without bias
AI reflects the biases in its training data. Data from unequal systems produces AI that perpetuates those inequalities.
AI understands context — it knows what you mean, not just what you said
AI processes the text you give it. It has no understanding of your organisation, your situation, or what you're trying to achieve beyond the words in your prompt.
AI can take responsibility — "the algorithm decided"
Under South African law and internationally, accountability for AI decisions rests with the people and organisations that deploy them — not the software.
Confident output = correct output
AI outputs are expressed with the same tone and style whether they are accurate or completely fabricated. Confidence is not a signal of accuracy.
You've seen what AI is, what it isn't, and the four types you're most likely to encounter. Now bring it to your world.
Answer these three questions about your own workplace:
Use general descriptions only. Do not include personal information (names, ID numbers, contact details, salary, medical or disciplinary history).
Generative = creates content
Predictive/Decision = scores, ranks, or classifies
Computer Vision = interprets images or video
Agentic = plans and takes multiple steps toward a goal
Not sure = describe what it does