1
What AI actually is
About 15 minutes
Ordinary software follows rules a person wrote down. AI is a broad name for software that does tasks we normally link with thinking: recognising a face, translating a sentence, suggesting a next word.
Some AI still runs on hand-written rules and search, like a route planner. Most of what makes headlines today is machine learning, where the system finds patterns in examples instead of being told the rules.
A useful habit: when you hear "AI", ask what the system does, what it was built from, and how anyone checked that it works.
Remember- AI is a field; machine learning is one approach inside it.
- A system that sounds human is not proof that it understands.
- Always ask what task it does and what evidence shows it works.
2
How machines learn from data
About 20 minutes
A machine learning system is shown many examples and adjusts itself until its guesses get better. That adjusting is called training. Using the trained system on something new is called inference.
The system can only learn what its examples contain. If the examples are narrow, old or one-sided, the system will be too. A model can also memorise its examples and fail on anything different, which is why it is tested on data it has never seen.
Good results on familiar data tell you little. What matters is how it does on new cases that look like the real world.
Remember- Training uses examples; inference uses the finished model.
- Output quality is capped by data quality.
- Test on new data, not on what the system already saw.
3
Models and their limits
About 20 minutes
A model is the saved result of training: a large set of numbers that turns an input into an output. Chat assistants are language models. They predict likely next words, one piece at a time, using patterns from huge amounts of text.
That is why they can be fluent and wrong at the same time. They are built to produce plausible text, not to look facts up. Newer tools add search and documents to help, but the check still belongs to you.
Treat a model like a quick, well-read assistant who sometimes guesses with total confidence.
Remember- A model turns inputs into outputs using learned patterns.
- Fluent does not mean correct.
- Verify anything that matters before you rely on it.
4
Asking AI well
About 20 minutes
The words you give an AI tool are called a prompt. Vague prompts get generic answers. Good prompts say who the answer is for, what you need, what to use, and what form it should take.
Give it the facts it needs, because it cannot see your situation. Ask for the steps or the sources behind an answer. If the first answer is off, say what is wrong and try again instead of starting over.
Never paste in anything private that you would not want stored by a third party.
Remember- State the goal, the audience, the facts and the format.
- Iterate: correct the answer, do not just accept it.
- Keep private details out of prompts.
5
Checking AI claims and outputs
About 20 minutes
AI is surrounded by big claims. A claim is only as good as its evidence: who measured it, on what, and compared to what? A single impressive demo is not a test.
AI tools also make things up, including names, numbers and citations that look real. The risk is highest where you cannot easily check, and where a mistake is costly: health, money, law, news.
A simple routine: find the claim, find the source, check one independent source, and note what is still uncertain.
Remember- Demos are not evidence; ask how it was tested.
- AI can invent facts and sources that look real.
- Verify more when the stakes are higher.
6
Bias and fairness
About 20 minutes
AI learns from human-made data, so it can pick up human-made unfairness. A hiring tool trained on past hires may copy past habits. A face tool tested mostly on one group may fail on others.
Bias is not always intentional, and it is not fixed by removing one obvious field, because other data can stand in for it. It shows up when you check results for different groups, not when you look at the average.
Fairness is a choice about trade-offs, so someone accountable has to make it and explain it.
Remember- Biased data produces biased results.
- Check outcomes for different groups, not just the average.
- A person stays accountable for what the system decides.
7
Privacy and who owns what
About 20 minutes
When you type into an AI tool, your words may be stored, reviewed or used to improve the service. Read the settings. Leave out passwords, health details, customer data and anything covered by an agreement.
Copyright around AI is unsettled. Tools are trained on large amounts of existing work, and the rules on who owns generated output differ by country and by tool. For anything you publish or sell, check the tool's terms and keep a record of how you made it.
When in doubt, use less personal data, and say when AI helped.
Remember- Assume inputs can be stored; protect personal data.
- Ownership of AI output is not simple; read the terms.
- Be open about where AI helped.
8
Using AI safely, and what comes next
About 20 minutes
AI helps most when a person stays in the loop: you decide what to ask, you review what comes back, and you own the result. The more serious the task, the more careful the review should be.
Risks to watch include scams made with AI voices and images, over-trusting a tool, and handing it jobs it cannot do reliably. Rules and norms are still forming, so expect change.
The most useful skill is not any one tool. It is knowing how to learn new ones, check them, and decide when not to use them.
Remember- Keep a human responsible for important decisions.
- Expect AI-made scams; verify surprising requests another way.
- Keep learning; tools change, judgement lasts.
Ready for the exam?
25 questions, 80% to pass. If you miss, review and try again after a short break. You need a free account so we can save your result.
Take the exam and get my certificate