The words of AI

Abstraction means keeping the details that matter for a purpose and hiding the rest.

Accountability means a named person or organisation can be asked to explain and put it right. If everyone can point elsewhere, there is none.

Accuracy = right ÷ total. — High accuracy still means some mistakes.

A pattern written as a rule is a formula. — A rule a machine can follow step by step is an algorithm — a word that is a man's name, as you learned in Class 6.

Artificial Intelligence (AI) is a computer doing a task that usually needs human intelligence — recognising a face, understanding speech, making a choice.

Automation follows fixed rules a person wrote — the same thing every time. A microwave, a washing-machine program, a traffic light on a timer.

Bag of words: a vocabulary of every unique word, and a document vector counting each word in each document. Word order is ignored.

Bias means results are unfairly tilted — better for some people, places or things than others.

A chatbot holds a conversation by text or voice — one of NLP's commonest uses.

Classification puts each item into one of a fixed set of groups — A/B/C, spam/not spam, healthy/sick.

Clustering groups items that are similar, with no labels given in advance.

Computer vision helps machines understand images and video.

Consent means the people agreed, knowing the use. Posting a photo isn't agreeing to train a model.

Convolution: multiply the image and the kernel element by element, add it all up, slide one step, repeat.

Data is raw facts we collect. Organised so that it tells us something, it becomes information.

Data literacy is being able to read data, work with it, analyse it, and argue with it. CBSE's Data Literacy Process Framework runs these as a loop.

A dataset is a collection of data about one topic.

A bug is a mistake in a program. Debugging is finding and fixing it.

A decision tree is a chain of yes/no questions that leads to a decision.

Decomposition means breaking a big problem into small parts you can manage.

A deepfake is generated video, audio or an image of a real person doing or saying what they never did.

A digital footprint is the trail you leave when you use the internet.

Error = prediction − real value. — Every prediction has some; good models keep it small.

Ethics means knowing right from wrong and choosing right. Computer ethics is that, for computers, phones and the internet.

F1 = 2 × P × R ÷ (P + R). One balanced number when both matter.

A feature is one measurable thing, defined so precisely that two people would collect it the same way.

Generative AI produces text, images and audio. A classifier asks which class?; this asks what comes next?

A hallucination is a confident, fluent statement that is simply false. It comes from how the model works; it can't be switched off.

Human in the loop: a person checks — and can change — an AI's suggestion before it becomes a decision.

Labelled data teaches it: examples where the right answer (the label) is known. Learning from them is training.

A loop tells a computer to do the same steps again and again — in Scratch, repeat — so you write the unit once.

Machine Learning (ML) is a machine learning from data instead of rules a person wrote.

Mean = total ÷ count · median = the middle value · mode = the commonest.

Misinformation is false information that spreads. Most people forwarding it aren't lying — they just never checked.

The model is what training makes: the part that does the job afterwards. In Part 1 your model told an apple from a pen.

NLP (natural language processing) helps machines understand and produce language — text and speech.

A neural network is layers of neurons: an input layer, hidden layers that do the working, an output layer that answers.

Never test on training data. — A model can remember the training set and answer it perfectly. That is overfitting: memory that looks like skill.

A pattern is something that repeats in a way we can notice.

Personal data — name, phone, address, face, date of birth — points to one person. Leave it out unless truly needed.

A pixel is a picture element: the smallest unit of a digital image. Every photo is a grid of them.

Precision is how exact or detailed a measurement is.

A prediction is a guess that uses a pattern from earlier data.

Privacy: who may use this, for what, and did they agree? Security: what stops someone else getting it?

Probability = favourable outcomes ÷ possible outcomes — a fraction, a decimal or a percentage.

Recall = TP ÷ (TP + FN). How much of what was real it caught. Matters when a miss is costly.

Regression predicts a number — temperature, price, marks — not a group.

Reinforcement learning learns by trial and error, from reward and penalty.

Sentiment analysis finds the feeling in text — positive, negative or neutral.

Structured data sits in rows and columns — easy to store, search, sort and analyse.

Supervised learning uses labelled data: every example comes with its answer.

TF-IDF = TF × log(N ÷ DF), log base 10. High when a word is common in one document and rare in the rest.

A token is a piece of text. Our place names cost more tokens than English words the same length.

Training is the learning — showing the machine data until it picks up the pattern.

Training data is labelled — every example carries its answer.

Transparency: people are told when AI is used on them, and how it decides, simply.

Unstructured data — images, video, audio, posts — has no fixed shape. Richer, but harder to analyse.

Unsupervised learning uses unlabelled data and finds the patterns on its own.

Split it three ways: training (learn) · validation (check and improve while training) · test (the final check, on data never seen).

Text inside quotes is printed exactly as you typed it. A name without quotes is a variable — a box holding whatever the user typed.