AI Skills Academy · Unit 1 of 8
Unit 1: What AI really is
How AI learns from examples instead of rules, the kinds of AI you already use, and what it is good and bad at.
Opens the first lesson of this unit in AI Skills Academy. Every lesson is open from the start.
Lesson 1: AI is software that learns patterns
Goal: Explain the difference between rule-based software and machine learning.
Most software follows rules a programmer wrote by hand: "if the order is over $50, shipping is free." Artificial intelligence (AI) is software that does tasks we used to think needed human judgment, like understanding speech, translating, recognizing faces or writing. Most modern AI gets there by learning patterns from examples instead of following hand-written rules.
Take a spam filter. Writing rules by hand ("block any email containing WINNER") breaks fast: spammers just write W1NNER. Machine learning takes a different route. You show a program thousands of emails that people have already labeled "spam" or "not spam", and it adjusts itself, a tiny bit at a time, until its guesses match the labels. What it ends up with is a model: the learned pattern, saved so it can be used on emails it has never seen.
Worked example: rules versus learning
- Rule-based: a programmer writes
if subject contains "WINNER" then spam. A spammer writes "W1NNER" and gets straight through. - Learned: a model trained on 50,000 labeled emails weighs hundreds of clues at once: the sender, the links, the wording, the time sent.
- When spammers change tactics, people label new examples and the model is retrained. Nobody has to rewrite every rule.
Learned models handle messy, changing situations far better than hand-written rules. That is why most AI today is built this way.
- AI (artificial intelligence)
- Software that performs tasks that normally need human judgment, such as recognizing images, understanding language or making predictions.
- Machine learning
- A way of building AI where the program learns patterns from examples instead of being given every rule.
- Training data
- The examples a model learns from. Its quality decides how good the model can be.
- Model
- The learned pattern itself, saved as a very large set of numbers, that turns new inputs into outputs.
The catch: a model is only as good as its examples. If its training emails never included a new kind of scam, it has never seen that pattern, and it can miss it. That one idea explains a lot of AI mistakes.
Remember: AI learns from examples. To predict what an AI will do well or badly, ask what it learned from.
Lesson 2: Kinds of AI you already use
Goal: Tell predictive AI from generative AI, with examples from daily life.
You probably use AI dozens of times a day without noticing. It helps to sort it into two big families.
Predictive AI looks at data and makes a judgment: it recommends (the next video), classifies (is this email spam, is this my face), detects (unusual card spending) or forecasts (when you will arrive). Its output is a choice, a label or a number.
Generative AI creates new content: text, images, audio, video or code. Chatbots that write and explain, image tools that turn a description into a picture, and voice tools that read text aloud are all generative.
| What it does | Family | Everyday example |
|---|---|---|
| Suggests the next song | Predictive | Music app recommendations |
| Unlocks your phone with your face | Predictive | Face recognition |
| Estimates your arrival time | Predictive | Map apps |
| Writes a draft email | Generative | AI writing assistants |
| Makes a picture from a description | Generative | Image generators |
| Turns speech into text | Predictive | Voice typing (it predicts the most likely words you said) |
Today's AI systems can be very capable at the tasks they were built and trained for, and surprisingly poor just outside them. A chatbot can explain algebra clearly and then miscount the letters in a word. Knowing which family a tool belongs to, and what it was built for, tells you how much to trust it.
- Predictive AI
- AI that turns data into a judgment: a recommendation, a label, a score or a forecast.
- Generative AI
- AI that creates new content such as text, images, audio, video or code.
Remember: Predictive AI chooses or scores. Generative AI creates.
Lesson 3: What AI is good and bad at
Goal: Predict when an AI answer deserves trust and when it needs checking.
AI is strongest where there are lots of patterns to learn and where a "pretty good" answer is useful: drafting, summarizing, translating, brainstorming, explaining a topic in simpler words, spotting patterns in large amounts of data.
It is weakest, or riskiest, in a few predictable places:
- Recent events. A model learns from data collected up to a date, its knowledge cutoff. Unless the app lets it search the web, it does not know what happened after that.
- Exact facts it rarely saw. Small towns, people who are not famous, exact quotes, page numbers and statistics are easy to get wrong.
- Knowing when it is wrong. A wrong answer can sound exactly as confident as a right one.
- Your private situation. It does not know your medical history, your bank balance or what your teacher actually asked unless you tell it, and some of that you should not tell it.
- Decisions that need judgment and accountability, like medical, legal, money or safety choices.
Worked example: the knowledge cutoff
You ask a chatbot with no web search: "Who won last night's game?"
- Its training data stops at its cutoff date, so "last night" is outside anything it learned.
- A good assistant says it cannot know. A weaker one may produce a confident, made-up score, because a plausible answer is what its training rewards.
- Better: use a tool that shows its sources, or check a sports site yourself.
The question to ask is always: could this AI actually know this?
Remember: Before trusting an answer, ask: could this AI actually know this, and how would I check?
Lesson 4: You are the pilot
Goal: Use the human-in-the-loop pattern and know when AI should not decide.
The people who get the most out of AI treat it like a fast, talented assistant who sometimes makes things up. They stay in charge. This is called keeping a human in the loop.
A simple way to remember it is the AI sandwich: a human starts it, the AI does the middle, and a human finishes it.
- You start: decide the goal and give a clear request with the facts it needs.
- AI drafts: it produces options, a draft or an analysis.
- You finish: check facts, fix errors, make it sound like you and decide what to actually use.
You are responsible for what you send, submit or act on, even if AI wrote it. "The AI said so" is never a defense with a teacher, a boss or a customer.
Watch out: Some decisions should not be handed to AI at all: medical symptoms and medicines, legal problems, big money decisions and anything about safety. AI can help you understand the topic and prepare questions, but the decision belongs with you and a qualified person, like a doctor, a lawyer or a trusted adult.
Worked example: the AI sandwich in action
You need to email your soccer coach that you will miss Saturday's practice.
- You start: "Write a short, polite email to my coach. I will miss Saturday's practice for my cousin's wedding. I will make up the conditioning drills on Sunday."
- AI drafts: a friendly four-sentence email.
- You finish: you notice it says "Saturday the 14th" when practice is on the 15th, fix the date, cut one cheesy line and send it from your own account.
Two minutes saved, and the one factual mistake was caught because a human checked.
Remember: Human starts, AI drafts, human finishes. You own the result.
Practise it in AI Skills Academy
Each lesson ends with two or three questions that check you can use the idea, not just recognise it. Wrong answers explain why and come back in Review. The unit ends with a 5-question quiz; pass it with 80% or more to master the unit.
Every lesson is free: no ads, no lives and no paywall.