You played three games today, and each one ran a different kind of AI. Rock-Paper-Scissors started beating you by round 12. Semantris scored the word "river" closer to "ocean" than "blue." Quick Draw named your sketch in three seconds. Which was which — and how did each one manage it?
That mystery opens the book. Three ideas hold the whole series together: Data (DS), Computer Vision (CV), and Natural Language Processing (NLP) — the three domains of AI. Every game you played this morning belongs to one of them.
Every AI tool in this book runs in one of these three, or a mix. The mystery comes first, the explanation second. By the end of this chapter, you will name the domain behind any AI tool you meet — and say why.
You already know what AI is — a system doing tasks that normally need human intelligence. It recognises an object, understands a sentence, decides from data. Class 9 sharpens that same question.
The question moves from WHAT the system does to HOW it does it. Did a person write a rule for every case? Or did the system find the rule itself, by studying data?
Ask that question of any AI system, and the answer sorts it. The rule is either a human-written list, or one the system found in the data. That single distinction splits every AI system you will meet into two families.
One question divides every AI system into two families: how does it know what to do? A rule-based system follows instructions a person wrote — an expert spells out every condition and the action to take. "IF it rains, THEN take an umbrella" is a rule someone set down in advance.
A learning-based system works differently. It finds the rule itself, by studying many labelled examples. No human writes the rule down anywhere — the data shapes the rule, not a person.
Neither family wins outright. A rule-based system is predictable and easy to check, because you can read every rule it follows. A learning-based system handles complexity no person could fully spell out. But you cannot always say exactly why it reached its decision. The three games you played this morning are all learning-based. Each one improved by seeing data, not by someone rewriting its rules.
Both families, rule-based and learning-based, operate inside three domains of AI. The domain is decided by one thing: what TYPE of data goes in.
Data, or Data Statistics (DS), reads values sitting in rows and columns — numbers and categories. It predicts a value or makes a decision. Computer Vision (CV) reads pixels, from images or video, to classify or detect what is shown. Natural Language Processing (NLP) reads language, in text or speech, to understand or generate it.
The domain question asks what type of data an AI works with — never how clever it is, or how it was built. A medical scan-reader is CV because its input is an image. A spam filter is NLP because its input is text. A loan-default predictor is DS because its input is a table of numbers.
Most real systems combine two domains or more. When they do, each subsystem still belongs to its own domain. The mixing does not erase the question — it just means you ask it more than once.
"Every AI system belongs to just one domain."
Most production systems cross domain lines. A voice assistant transcribes speech — that is NLP — reads your location off a sensor table — that is DS — and may read a camera feed — that is CV. The three domains are an analytical tool for understanding AI, not a physical wall between products. Saying "this system is NLP" names its primary input type. It does not mean no other domain is involved.
Look at the bazaar street you mapped back in Class 7. Every sector uses all three domains, but each one leans on a dominant one. Naming the sector tells you which domain is most likely primary.
Healthcare AI is CV-dominant: scan reading, cell counting, spotting an anomaly in an image. Finance and logistics are DS-dominant: risk scoring, route optimisation, demand forecasting from tables of numbers. Customer service and education are NLP-dominant: chatbots, content generation, essay assessment.
Every sector uses all three, but each has a dominant one. The same street, the same four shops — now labelled by what kind of AI actually runs inside each one.
Here the mystery resolves. Each game you played was a domain probe. A domain probe finds out which AI runs a system — by asking one question: what type of data is actually going in?
Rock-Paper-Scissors reads a sequence of your past moves — a number-pattern building up in a table over time. That is DS. Semantris scores how close the word you typed sits to a target word in meaning. The input is language, so that is NLP. Quick Draw classifies the live stream of pixels in your sketch — the input is visual, so that is CV.
Name the data type, and you have named the domain. The method generalises to any AI tool you meet after this chapter — not just these three games.
Weaker. By topic. "Rock-Paper-Scissors is a strategy game," Dhruv says, "so its AI must be about logic — that's not really a data domain." He is reading the domain off the game's theme, and the theme tells him nothing true about what the AI actually processes.
Stronger. By data type. Nila checks what actually goes into this AI: a sequence of past moves — rock, paper, or scissors, three categories building up over time in a table. That is structured data, so the domain is DS. She adds the test that settles it: let players speak their move instead, and the listening subsystem becomes NLP — same game, same rules, a different domain, because the data type changed.
"The game tells you the domain because of its topic — Rock-Paper-Scissors is about strategy, so it must be logic, not data."
The domain has nothing to do with a game's theme; it depends only on the type of data the AI actually processes. Rock-Paper-Scissors is DS because the input is a sequence of past moves sitting in a table. Build a version where players speak their move and the AI listens instead, and that subsystem becomes NLP — even though the game itself is unchanged. Data type decides the domain. Topic never does.
Now the most important idea in this chapter — the one every later chapter comes back to. Quick Draw guessed "house" from your sketch in three seconds, and it felt like the machine understood exactly what a house is.
It did not. Quick Draw has no concept of a house at all. It matched the pattern in your pixels to a cluster of similar pixel patterns from the sketches it trained on — nothing more.
Semantris does the same trick in a different domain. It ranked "river" closer to "ocean" than to "blue," and that felt like a sense of meaning. It has none. It computed a distance between word vectors, built from how often words sit near each other across huge amounts of text.
CV does not see. NLP does not understand. DS does not reason about why its numbers point where they do — each one found a statistical pattern, and stopped there. Correct output is not evidence of understanding. It is evidence that the training pattern matched the input pattern closely enough.
"If the AI gives the correct answer, it must understand the question."
Correct output means the pattern matched — not that the AI understood anything. Quick Draw guessed "house" because your pixels matched a cluster of training sketches labelled "house"; it holds no concept of a house. Semantris ranks words by the distance between vectors trained on co-occurrence statistics; it holds no sense of meaning. Understanding needs a model of the world. Pattern-matching needs only that the training data once held something similar.
There is a name for what this chapter just built in you: AI Readiness, the threshold idea for this whole unit.
At Grade 9, it means three things you can actually do. Identify the AI when you meet it in daily life. Name the domain it runs in — DS, CV, or NLP. Explain, in plain language, what that AI can do, and what it cannot.
A student who can do all three with a tool they have never seen before has crossed the threshold. A student who can only list AI products by name has product awareness, not readiness. The threshold is to EXPLAIN, not to ENUMERATE.
"AI Readiness means knowing a lot of AI tools by name."
Listing product names — GPT, Siri, Gemini — is product awareness, not AI Readiness. A student who names twenty products but cannot say which domain any one of them operates in, or what type of data each processes, has awareness without readiness. The real threshold is the ability to explain a tool you have never seen before: identify it, name its domain, and say what it can and cannot do.