Data and AI in the classroom: a teacher's guide

Algorithm

In plain English

A set of step-by-step instructions for solving a problem or finishing a task. A computer follows it exactly, in the order given.

Think of it like this

A recipe, or the steps for long division. Follow them in order and you get the answer.Where that stops working: Some modern AI systems are not written as steps by a person. They find their own rule from examples, which is what the next few entries are about.

At work

Data jobs involve writing or choosing algorithms, and deciding which problem to solve with one is as important as writing the code.

Artificial intelligence (AI)

Also called: AI

In plain English

A broad label for computer systems that do things which normally seem to need human intelligence, such as understanding speech, recognising pictures or writing text. It covers many different methods.

Think of it like this

'Sport' is a label for football, swimming and chess boxing. AI is a label for many different techniques in the same way.Where that stops working: Nobody agrees exactly where the edge of the label is, and the word is often used for things that are just ordinary software.

At work

A job with AI in the title is usually about building, checking or running a particular system, rather than anything as grand as the film version.

Automated decision

Also called: automated decision-making, ADM

In plain English

A decision about a person made solely by a computer system with no human involved. The ICO's guidance is about such decisions where they have a legal or similarly significant effect, and the safeguards people should have.

Think of it like this

A system that decides automatically which students get a place on a course, with no person looking at it.Where that stops working: This is general information, and the ICO's page is draft guidance under consultation.

At work

Governance and operations roles check that decisions made by systems can be explained, challenged and corrected.

Average (mean, median, mode)

Also called: mean, median, mode

In plain English

Three ways of describing the middle of some numbers. The mean adds them up and divides, the median is the middle value in order and the mode is the most common.

Think of it like this

Three students at the bus stop earn £5 each from a Saturday job and one earns £100. The mean is about £29, the median is £5. Which one is more honest about a typical student?Where that stops working: No single average describes everyone, which is one reason a headline average can hide how different groups do.

At work

Choosing which average to report is a real decision, and a good analyst says which one they used.

Bias

Also called: algorithmic bias

In plain English

In data and AI, a system that works worse, or gives systematically different results, for some groups of people. It often comes from the examples the system learned from or from choices made in building it.

Think of it like this

A school quiz written entirely by students from one friendship group will probably suit that friendship group best.Where that stops working: In everyday life bias suggests someone meant it. A system can be biased with nobody meaning anything.

At work

The Career Matrix has a whole area of work, Data Strategy, Governance and Ethics, about using data responsibly, which includes looking for this.

Causation

In plain English

When one thing really makes another happen. Much harder to show than correlation, and usually needs an experiment or very careful design.

Think of it like this

Pressing the switch makes the projector come on. Doing it again and again with the same result is how you know.

At work

Applied data scientists run experiments, such as splitting users into groups, to move from 'these go together' to 'this caused that'.

Confounder

Also called: third factor, lurking variable

In plain English

A hidden third thing that causes two other things to move together, so they look related when they are not.

Think of it like this

Rain makes more umbrellas sell and more students arrive late. Umbrellas do not cause lateness; the weather causes both.

At work

Spotting possible confounders is one of the first things an analyst is trained to do.

Correlation

In plain English

Two things that tend to rise and fall together. It is a pattern in the numbers, and says nothing yet about why.

Think of it like this

Students who sit near the radiator tend to be sleepier. The two go together; that does not mean the radiator is the whole reason.Where that stops working: A strong correlation can still be a coincidence or the work of a third thing.

At work

Analysts use correlation to decide where to look, not to prove a cause.

Critical National Infrastructure (CNI)

Also called: CNI

In plain English

The systems and facilities a country relies on so much that their loss would be serious, such as energy and water. In 2024 the government announced it would designate data centres as CNI.

Think of it like this

The school's power supply: nobody thinks about it until it fails, and then nothing works.

At work

Data centre engineers and technicians keep this running, and security professionals help protect it.

Dashboard

In plain English

A single screen of charts and key figures that lets someone keep an eye on how something is going.

Think of it like this

The attendance board in the office: one glance tells you which classes need chasing.Where that stops working: A dashboard shows what someone decided to measure and nothing else. The Make the same data say three things activity is a warning about that.

At work

Analysts build dashboards. The best ones also say what the dashboard cannot tell you.

Data cleaning

Also called: data wrangling, data preparation

In plain English

Finding and fixing errors in data before using it: duplicates, wrong units, impossible values, inconsistent labels and gaps.

Think of it like this

Marking your own homework for silly mistakes before you hand it in.Where that stops working: Cleaning involves judgement. The Clean the mess activity has a 262 cm height that nobody can safely correct without asking.

At work

Data engineers and analysts do this work. It is not glamorous, and everything after it depends on it.

Data pipeline

In plain English

The automated route data takes from where it is collected, through cleaning and checking, to where it is used.

Think of it like this

Homework moving from your bag, to the teacher's pile, to the mark book, to the report, without anyone retyping it.Where that stops working: A mistake early in the pipeline travels all the way to the end, so the checking has to happen along the way.

At work

Data engineers build and maintain pipelines. The Matrix lists the role in its first area of work.

Database

In plain English

An organised store of data that lets you add, find and change information quickly and reliably.

Think of it like this

The school's management system, where one search finds a student, their timetable and their parent's phone number.

At work

Database specialists and data engineers keep these running, and the Career Matrix lists them in the Data Foundations and Engineering area.

Dataset

In plain English

A collection of data organised for a purpose, often rows (one per person, event or item) and columns (one per thing you measured).

Think of it like this

The class register: one row per student, one column per thing you record.

At work

Nearly every project starts by asking which datasets exist, who owns them and whether they can be used.

Generative AI

Also called: gen AI

In plain English

AI that produces new content, such as text, pictures or music, based on patterns it learned from a lot of examples.

Think of it like this

A student who has read thousands of stories and can now write one of their own in the same style, without having to copy any single one.Where that stops working: The new content can look convincing and be wrong. See hallucination.

At work

Anyone who can judge where generative AI helps and where it does not is useful to an employer.

Hallucination

In plain English

When an AI tool produces false or nonsensical information and presents it as fact. The DfE describes it in just those words.

Think of it like this

A student who has not done the reading and bluffs, in a very confident voice.Where that stops working: A bluffing student knows they are bluffing. A model has no idea.

At work

Checking the output of AI tools is part of nearly every job that uses them. The DfE says any content produced requires critical judgement to check for appropriateness and accuracy.

Large language model (LLM)

Also called: LLM, chatbot

In plain English

The NCSC describes it as an algorithm trained on a large amount of text that analyses the relationships between words and turns them into a probability model, which it uses to answer. In short, it works out likely continuations of the text it is given.

Think of it like this

The Predict the next word activity is a tiny, readable version of the same idea.Where that stops working: The activity counts word pairs in a few dozen sentences. Real models are trained on vastly more text and use far more powerful methods. The shared idea is choosing a likely next word.

At work

People who build, test and secure these systems work in machine learning, operations and security roles. The NCSC has published advice on using them safely.

Machine learning

Also called: ML

In plain English

A way of making software where, instead of writing the rules by hand, you show a program lots of examples and let it work out the pattern for itself.

Think of it like this

Teaching a new student to spot a good sentence by showing them hundreds of good and bad ones, rather than handing them a rulebook.Where that stops working: A person still decides which examples to show and how to judge whether the student has got it right, and that is where many of the problems start.

At work

Machine learning engineers build and train these systems. The Career Matrix lists a Level 6 degree apprenticeship route into it.

Model

In plain English

What machine learning produces once it has learned from data: a pattern-spotting rule you can use on new cases.

Think of it like this

The 'feel' a football coach has for a good striker after watching thousands of matches. It is not written down anywhere, but it can be used on a player they have never seen.Where that stops working: A coach can explain a hunch. A model often cannot explain why it decided what it did.

At work

Data scientists and machine learning engineers build models. Others check them, run them and keep an eye on them once they are in use.

Neural network

In plain English

A kind of model made of very many small, connected calculations whose strengths are adjusted during training. The name is loosely inspired by the brain, but it is maths, not a brain.

Think of it like this

A huge number of tiny dials. Training is a long process of nudging every dial until the answers improve.Where that stops working: Nobody can read the dials and say what each one means, which is why explaining a decision is hard.

At work

Machine learning engineers choose, train and test networks.

Outlier

In plain English

A value that sits far away from the rest. It might be an error, or it might be the most interesting thing in the dataset.

Think of it like this

The one student who scored 100% when everyone else was between 40 and 60. A mistake, or a genius?

At work

Analysts look at outliers before deciding what to do with them. Removing them without saying so is a classic way to mislead.

Overfitting

In plain English

When a model learns its training examples so closely, quirks and mistakes included, that it does badly on new cases.

Think of it like this

Learning last year's past paper answers word for word and then freezing when the questions are changed slightly.Where that stops working: A student can tell they are stuck. A model just confidently gives the wrong answer.

At work

Testing on data the model has not seen is how teams catch it.

Personal data

In plain English

Any information about a living person who could be identified from it. In schools it includes names, photos, grades and anything that points to a particular student.

Think of it like this

Everything in a student's file, and a photo from the school trip.Where that stops working: The details of the law are for your school's data protection officer. The DfE's advice is to keep personal data out of generative AI tools.

At work

Anyone handling data needs to know what counts, who can see it and why it is held. Governance roles in the Matrix focus on this.

Prompt

In plain English

The text you give an AI tool to say what you want from it.

Think of it like this

The question on an exam paper. A vague question gets a vague answer, and a different wording can get a very different one.

At work

Writing good prompts is a useful skill, but it is not a substitute for knowing enough about the topic to check the answer.

Sample

In plain English

The group you actually measure, used to learn about a larger group you cannot measure completely.

Think of it like this

Tasting one spoonful of soup to judge the whole pot. It only works if you stir it first.Where that stops working: If the sample is not representative, you learn about the sample and not the pot. The Teach the machine game is that problem in miniature.

At work

Whether a sample represents the people a system will be used on is one of the first questions a good team asks.

SQL

In plain English

A language for asking questions of a database, such as 'show me every order over £50 from last March'.

Think of it like this

Asking the librarian for every book on volcanoes published after 2010, in a form the library system understands exactly.Where that stops working: It only answers what you ask. A badly worded question gets a confident wrong answer.

At work

Analysts use it to get answers out of databases.

Structured and unstructured data

In plain English

Structured data fits neatly into rows and columns, like a register. Unstructured data does not, like photos, recordings or free-text comments.

Think of it like this

The register is structured. The comments teachers write on reports are not.

At work

Much of today's AI work is about making use of unstructured data, and someone still has to organise and check it.

Test data

Also called: test set

In plain English

Examples kept back and never shown to the model while it learns, then used to find out how well it really works.

Think of it like this

The mock exam: questions you have not seen, used to check what you actually know.Where that stops working: A mock only tells you about the topics it covers. A test set that leaves out some people tells you nothing about those people.

At work

Testing is where a team finds out who a system fails. The Teach the machine game shows why that has to be checked group by group.

Training data

Also called: training set

In plain English

The examples a model learns from. What goes in shapes what comes out.

Think of it like this

The textbook a student revises from. If the textbook leaves out a topic, the student will not know it however clever they are.Where that stops working: A student can notice a gap and go looking. A model cannot, unless a person spots it first.

At work

Someone has to collect, check and label training data, and the quality of the whole system depends on it.