AI literacy for teachers and students · Keep Humans in the Loop

The AI Literacy Periodic Table

AI literacy means more than knowing the words. This table helps students and staff understand AI, question it, and decide when a human must stay in charge.

Word of the week · 16 Nov · Safe theme

Memory

What do you think an AI tool remembers about you? How could an adult help you check?

Open it
View by

Drag the slider, or press Play, to watch the words arrive. Dates show when each term was in use in AI, not necessarily the very first time anyone said it; "about" means it came into use gradually. Some words, like Bias and Privacy, are much older: their year is when they took on their AI meaning. Know an earlier date? Tell us.

Print the poster

Foundations Smart theme In use since 1955

Artificial Intelligence

Computer systems, built and used by people, that do tasks we once thought only humans could do, such as translating text, recognising faces in photos or recommending what to watch next.

Classroom notes & resources

Issues it raises · choose one to read about it

Benchmark: AI Literacy

Element 1 of 142

Foundations The ideas everything else is built on.

Generative AI AI that makes new text, images, sound, video and code.

Predictive & Analytical AI AI that finds patterns in data to sort, forecast and recommend.

Autonomous AI Agents AI that plans and takes actions towards a goal.

Automation & Workflow AI AI and software that carry out tasks and processes.

Governance & Ethics Keeping AI fair, safe, lawful and accountable, and the ways people get around it.

Infrastructure & Ops The hardware, tools and methods that run AI.

Safe theme

Smart theme

Creative theme

Responsible theme

Future theme

Before 1950

1950s–70s

1980s–90s

2000–2014

2015–2021

2022–2023

2024 on

1 Ai Artificial Intelli­gence , Foundations 2 Ls Latent Space , Foundations 3 Rsi Recur­sive Self-Improve­ment , Foundations 4 Si Super­intelli­gence , Foundations 5 Agi Artificial General Intelli­gence , Foundations 6 Ml Machine Learning , Foundations 7 Dl Deep Learning , Foundations 8 Ag AI Agent , Autonomous AI Agents 9 Ac Agentic AI , Autonomous AI Agents 10 Au Auto­nomy , Autonomous AI Agents 11 Rp Robotic Process Auto­mation , Automation & Workflow AI 12 Ia Intelli­gent Auto­mation , Automation & Workflow AI 13 Hy Hyper­auto­mation , Automation & Workflow AI 14 Nn Neural Network , Foundations 15 Al Algorithm , Foundations 16 Pl Planning , Autonomous AI Agents 17 Me Memory , Autonomous AI Agents 18 Fn Function Calling , Autonomous AI Agents 19 Wo Workflow Orches­tration , Automation & Workflow AI 20 Tg Trigger , Automation & Workflow AI 21 En Event-Driven Auto­mation , Automation & Workflow AI 22 Md Model , Foundations 23 Tr Training , Foundations 24 Ll Large Language Model , Generative AI 25 Ga Gener­ative AI , Generative AI 26 Pr Prompt , Generative AI 27 Pe Prompt Engin­eering , Generative AI 28 Cw Context Window , Generative AI 29 Su Super­vised Learning , Predictive & Analytical AI 30 Un Un­super­vised Learning , Predictive & Analytical AI 31 Rr Regres­sion , Predictive & Analytical AI 32 Cl Classi­fication , Predictive & Analytical AI 33 Cu Cluster­ing , Predictive & Analytical AI 34 Ra ReAct , Autonomous AI Agents 35 Lp Agent Loop , Autonomous AI Agents 36 Sr Self-Reflec­tion , Autonomous AI Agents 37 Ap API Integ­ration , Automation & Workflow AI 38 Wh Webhook , Automation & Workflow AI 39 Lc Low-Code / No-Code , Automation & Workflow AI 40 If Inference , Foundations 41 Pm Para­meters , Foundations 42 Mm Multi­modal AI , Generative AI 43 Df Diffusion Model , Generative AI 44 Gn Gener­ative Adver­sarial Network , Generative AI 45 Ti Text-to-Image , Generative AI 46 Ts Text-to-Speech , Generative AI 47 Dt Decision Tree , Predictive & Analytical AI 48 Rf Random Forest , Predictive & Analytical AI 49 Gb Gradient Boosting , Predictive & Analytical AI 50 Fe Feature Engin­eering , Predictive & Analytical AI 51 Of Over­fitting , Predictive & Analytical AI 52 Mc Model Context Proto­col , Autonomous AI Agents 53 Aa Agent-to-Agent (A2A) , Autonomous AI Agents 54 Or Orches­trator , Autonomous AI Agents 55 Id Intelli­gent Docu­ment Process­ing , Automation & Workflow AI 56 Oc Optical Charac­ter Recog­nition , Automation & Workflow AI 57 Mi Process Mining , Automation & Workflow AI 58 Tk Token , Foundations 59 Em Embed­ding , Foundations 60 Ft Fine-Tuning , Generative AI 61 Rg Retriev­al-Augment­ed Gener­ation , Generative AI 62 Te Temper­ature , Generative AI 63 Ct Chain-of-Thought , Generative AI 64 Fs Few-Shot Learning , Generative AI 65 Fc Fore­casting , Predictive & Analytical AI 66 Tm Time Series , Predictive & Analytical AI 67 Ad Anomaly Detection , Predictive & Analytical AI 68 Pa Predic­tive Analy­tics , Predictive & Analytical AI 69 Ps Prescrip­tive Analy­tics , Predictive & Analytical AI 70 Ma Multi-Agent System , Autonomous AI Agents 71 Sw Agent Swarm , Autonomous AI Agents 72 Rl Rein­force­ment Learning , Autonomous AI Agents 73 Be Business Rules Engine , Automation & Workflow AI 74 St Straight-Through Process­ing , Automation & Workflow AI 75 Dw Digital Worker , Automation & Workflow AI 76 Tf Trans­former , Foundations 77 Fm Foun­dation Model , Foundations 78 Rm Reason­ing Model , Generative AI 79 Sd Synthetic Data , Generative AI 80 Ha Halluci­nation , Generative AI 81 Dk Deepfake , Generative AI 82 Vc Vibe Coding , Generative AI 83 Rc Recom­mend­ation Engine , Predictive & Analytical AI 84 Ch Churn Pre­diction , Predictive & Analytical AI 85 Sa Senti­ment Analysis , Predictive & Analytical AI 86 Np Natural Language Process­ing , Predictive & Analytical AI 87 Cv Com­puter Vision , Predictive & Analytical AI 88 Co Com­puter Use , Autonomous AI Agents 89 Br Browser Agent , Autonomous AI Agents 90 Ca Coding Agent , Autonomous AI Agents 91 Cb Chatbot , Automation & Workflow AI 92 Va Virtual Assistant , Automation & Workflow AI 93 Cp Copilot , Automation & Workflow AI 94 Ce Context Engin­eering , Generative AI 95 Gv Gener­ative Video , Generative AI 96 Sp AI Slop , Generative AI 97 2d 2D Design , Generative AI 98 3d 3D Design & CAD , Generative AI 99 Ge Gener­ative Engine Optim­isation , Predictive & Analytical AI 100 Rs Respon­sible AI , Governance & Ethics 101 Et AI Ethics , Governance & Ethics 102 Bi Bias , Governance & Ethics 103 Fa Fairness , Governance & Ethics 104 Xa Explain­ability , Governance & Ethics 105 Tn Trans­parency , Governance & Ethics 106 An Align­ment , Governance & Ethics 107 Gr Guard­rails , Governance & Ethics 108 Hi Human-in-the-Loop , Governance & Ethics 109 Rt Red Teaming , Governance & Ethics 110 Sf AI Safety , Governance & Ethics 111 Pv Privacy , Governance & Ethics 112 Dg Data Gover­nance , Governance & Ethics 113 Re AI Regu­lation , Governance & Ethics 114 Ip Copy­right & IP , Governance & Ethics 115 Aw AI Washing , Governance & Ethics 116 Sh Shadow AI , Governance & Ethics 117 Hp Human Premium , Governance & Ethics 118 Wm AI Water­marking , Governance & Ethics 119 So AI Sover­eignty , Governance & Ethics 120 Rv Reverse Engin­eering , Governance & Ethics 121 Mp Mechan­istic Interpret­ability , Governance & Ethics 122 As Activ­ation Steering , Governance & Ethics 123 Mx Model Extrac­tion , Governance & Ethics 124 Rd Refusal Direc­tion , Governance & Ethics 125 Ab Abli­ter­ation , Governance & Ethics 126 Uc Uncen­sored Model , Governance & Ethics 127 Jb Jail­breaking , Governance & Ethics 128 Gp GPU , Infrastructure & Ops 129 Tp TPU & Accel­er­ators , Infrastructure & Ops 130 Cd Cloud AI , Infrastructure & Ops 131 Ed Edge AI , Infrastructure & Ops 132 Mo MLOps , Infrastructure & Ops 133 Lo LLMOps , Infrastructure & Ops 134 Vd Vector Database , Infrastructure & Ops 135 Ms Model Serving , Infrastructure & Ops 136 Qz Quanti­sation , Infrastructure & Ops 137 Ds Distil­lation , Infrastructure & Ops 138 Ow Open-Weight Model , Infrastructure & Ops 139 Sm Small Language Model , Infrastructure & Ops 140 Ev Evals , Infrastructure & Ops 141 Sl Scaling Laws , Infrastructure & Ops 142 Dp Data Pipeline , Infrastructure & Ops

Left: foundations. Centre: generative and predictive AI. Right: autonomous agents and automation. Below: the governance and infrastructure that hold it all together. Numbers run left to right, top to bottom.

51 compounds · Stable or volatile?

Compound combinations

Elements combine into the AI systems people use every day. Each compound shows which elements a real system needs, where a human stays in the loop, and a question for your class to debate. Want to work them out yourself? Take the compound challenge below.

Stable A human stays in the loop. Volatile Risky without a human in the loop. Can you find the element that stabilises it?

Challenge on: compounds you haven't discovered yet are hidden. Each one gives you a clue: how many elements it needs, and one of them.

  • Undiscovered compound

    Clue: 3 elements, including Large Language Model (Ll).

    Looks like: A school helpdesk bot that answers staff questions straight from the policy handbook.

    Smart strand

    Stable Further education

    Grounded Assistant

    Smart strand

    Made from

    • Ll Large Language Model
    • Rg Retrieval-Augmented Generation
    • Vd Vector Database
    More about it

    A chatbot that looks up trusted documents before it answers, so its replies are based on real sources rather than memory alone.

    In real life
    A school helpdesk bot that answers staff questions straight from the policy handbook.
    Human in the loop
    A person keeps the documents up to date and spot-checks the answers.

    Debate it

    Should a school chatbot answer parents’ questions without a member of staff checking?

  • Undiscovered compound

    Clue: 3 elements, including Chatbot (Cb).

    Looks like: Asking a chatbot to explain photosynthesis in simpler words.

    Smart strand

    Stable Secondary

    General Chatbot

    Smart strand

    Made from

    • Cb Chatbot
    • Ll Large Language Model
    • Tf Transformer
    More about it

    A chat window on top of a large language model, built on the transformer. It predicts the most likely next words, which is why it can sound sure even when it is wrong.

    In real life
    Asking a chatbot to explain photosynthesis in simpler words.
    Human in the loop
    You check anything that matters against a trusted source before you rely on it.

    Debate it

    Should students have to say when they have used a chatbot for homework?

  • Undiscovered compound

    Clue: 3 elements, including Chatbot (Cb).

    Looks like: A tutor bot that offers hints and questions instead of full answers.

    Safe strand

    Stable Secondary

    Safe Study Buddy

    Safe strand

    Made from

    • Cb Chatbot
    • Ll Large Language Model
    • Gr Guardrails
    More about it

    An LLM chatbot with guardrails, designed to coach students rather than do the work for them.

    In real life
    A tutor bot that offers hints and questions instead of full answers.
    Human in the loop
    Teachers set the rules and review how it is being used.

    Debate it

    Should AI tutors give students answers, or only hints?

  • Undiscovered compound

    Clue: 2 elements, including Large Language Model (Ll).

    Looks like: An essay with references to books and articles that were never written.

    Smart strand

    Volatile Secondary

    Fake Citation

    Smart strand

    Made from

    • Ll Large Language Model
    • Ha Hallucination
    More about it

    A language model confidently inventing sources, quotes or facts that do not exist.

    In real life
    An essay with references to books and articles that were never written.
    Human in the loop
    Check every source yourself before you use it.

    Debate it

    When AI gets a fact wrong, who is responsible: the tool or the person who used it?

  • Undiscovered compound

    Clue: 2 elements, including Reasoning Model (Rm).

    Looks like: Getting a hint for the next line of an equation, rather than the answer.

    Smart strand

    Stable Secondary

    Step-by-Step Tutor

    Smart strand

    Made from

    • Rm Reasoning Model
    • Ct Chain-of-Thought
    More about it

    A reasoning model that writes out its working in steps before it answers, so you can check each step. The steps shown are not always how it really reached the answer.

    In real life
    Getting a hint for the next line of an equation, rather than the answer.
    Human in the loop
    You check each step and do the working yourself.

    Debate it

    If AI can show every step, does it matter whether you can do it alone?

  • Undiscovered compound

    Clue: 3 elements, including Large Language Model (Ll).

    Looks like: Searching a homework question and getting a summary before any links.

    Smart strand

    Stable Further education

    AI Search Summary

    Smart strand

    Made from

    • Ll Large Language Model
    • Rg Retrieval-Augmented Generation
    • Ge Generative Engine Optimisation
    More about it

    An AI-written answer at the top of search results, built from the pages it retrieves. Generative engine optimisation is how websites try to be part of it.

    In real life
    Searching a homework question and getting a summary before any links.
    Human in the loop
    You open the sources it used and check they say what the summary claims.

    Debate it

    If fewer people click through to websites, who will keep writing them?

  • Undiscovered compound

    Clue: 4 elements, including Transformer (Tf).

    Looks like: How a chatbot is built long before you type your first prompt.

    Smart strand

    Stable Further education

    LLM Factory

    Smart strand

    Made from

    • Tf Transformer
    • Tk Token
    • Tr Training
    • Gp GPU
    More about it

    The recipe for a large language model: a transformer trained on vast amounts of text tokens using huge numbers of GPUs.

    In real life
    How a chatbot is built long before you type your first prompt.
    Human in the loop
    People choose the training data and test the result.

    Debate it

    Should we have the right to know what data an AI was trained on?

  • Undiscovered compound

    Clue: 3 elements, including AI Agent (Ag).

    Looks like: An agent that drafts a school trip plan and booking, then waits for a teacher to approve it.

    Responsible strand

    Stable Secondary

    Supervised Agent

    Responsible strand

    Made from

    • Ag AI Agent
    • Fn Function Calling
    • Hi Human-in-the-Loop
    More about it

    An agent that uses tools to get work done, but pauses for human approval before anything important happens.

    In real life
    An agent that drafts a school trip plan and booking, then waits for a teacher to approve it.
    Human in the loop
    A human signs off every consequential action.

    Debate it

    Which decisions should an AI agent never make without asking first?

  • Undiscovered compound

    Clue: 3 elements, including AI Agent (Ag).

    Looks like: An agent with access to your email that replies to people and buys things on its own.

    Smart strand

    Volatile Secondary

    Unchecked Agent

    Smart strand

    Made from

    • Ag AI Agent
    • Fn Function Calling
    • Au Autonomy
    More about it

    An agent given tools and full autonomy, acting without anyone checking what it does.

    In real life
    An agent with access to your email that replies to people and buys things on its own.
    Human in the loop
    None, and that is the risk. Swap Autonomy (Au) for Human-in-the-Loop (Hi) to stabilise it.

    Debate it

    If an AI agent makes a costly mistake, who is responsible?

  • Undiscovered compound

    Clue: 4 elements, including AI Agent (Ag).

    Looks like: An assistant that manages your calendar, drafts emails and learns your routines.

    Safe strand

    Stable Further education

    Personal Assistant Agent

    Safe strand

    Made from

    • Ag AI Agent
    • Mc Model Context Protocol
    • Pl Planning
    • Me Memory
    More about it

    An agent connected to your apps that plans tasks and remembers your preferences over time.

    In real life
    An assistant that manages your calendar, drafts emails and learns your routines.
    Human in the loop
    You decide what it can access and what it is allowed to remember.

    Debate it

    Should an AI assistant remember things about you automatically, or only when you switch memory on?

  • Undiscovered compound

    Clue: 3 elements, including Computer Use (Co).

    Looks like: Comparing train tickets and filling in the booking form for you.

    Smart strand

    Stable Secondary

    Digital Errand Runner

    Smart strand

    Made from

    • Co Computer Use
    • Br Browser Agent
    • Ag AI Agent
    More about it

    An agent that operates a web browser like a person to complete online tasks.

    In real life
    Comparing train tickets and filling in the booking form for you.
    Human in the loop
    You confirm before it pays for or submits anything.

    Debate it

    Would you let an AI spend your money?

  • Undiscovered compound

    Clue: 3 elements, including Multi-Agent System (Ma).

    Looks like: One agent researches, another writes and a third fact-checks a report.

    Smart strand

    Stable Further education

    Agent Team

    Smart strand

    Made from

    • Ma Multi-Agent System
    • Or Orchestrator
    • Hi Human-in-the-Loop
    More about it

    Specialised agents coordinated by an orchestrator, passing work between each other.

    In real life
    One agent researches, another writes and a third fact-checks a report.
    Human in the loop
    A person sets the goal and signs off the final output.

    Debate it

    If agents check each other’s work, do we still need a human checker?

  • Undiscovered compound

    Clue: 3 elements, including Coding Agent (Ca).

    Looks like: Building a revision quiz app just by describing what it should do.

    Future strand

    Stable Secondary

    AI Pair Programmer

    Future strand

    Made from

    • Ca Coding Agent
    • Cp Copilot
    • Hi Human-in-the-Loop
    More about it

    A coding agent that builds software from plain-language requests, with tests to check the result.

    In real life
    Building a revision quiz app just by describing what it should do.
    Human in the loop
    A developer reviews, tests and understands the code before it ships.

    Debate it

    If AI writes the code, which coding skills still need to stay human?

  • Undiscovered compound

    Clue: 3 elements, including Diffusion Model (Df).

    Looks like: Creating poster artwork for the school play from a written description.

    Creative strand

    Stable Secondary

    AI Art Generator

    Creative strand

    Made from

    • Df Diffusion Model
    • Ti Text-to-Image
    • Ip Copyright & IP
    More about it

    A diffusion model that turns text prompts into images, raising real questions about who owns the result.

    In real life
    Creating poster artwork for the school play from a written description.
    Human in the loop
    A person directs, selects and edits the image, and credits the AI honestly.

    Debate it

    If AI made the picture from your prompt, who really made it?

  • Undiscovered compound

    Clue: 2 elements, including Text-to-Image (Ti).

    Looks like: A social media app adding an “AI-generated” label to a picture.

    Creative strand

    Stable Secondary

    Labelled AI Image

    Creative strand

    Made from

    • Ti Text-to-Image
    • Wm AI Watermarking
    More about it

    An AI-generated picture carrying a hidden watermark or attached content credentials, so tools can check later that it was AI-made. These signals can sometimes be stripped out.

    In real life
    A social media app adding an “AI-generated” label to a picture.
    Human in the loop
    Creators label their work honestly, and platforms check for the watermark.

    Debate it

    Should every AI-made image have to carry a label?

  • Undiscovered compound

    Clue: 3 elements, including Generative AI (Ga).

    Looks like: Harm: scam calls that imitate a family member asking urgently for money. Help: people losing their voice to illness recording it first, so they can keep speaking in their own voice.

    Safe strand

    Volatile Secondary

    Voice Clone

    Safe strand

    Made from

    • Ga Generative AI
    • Ts Text-to-Speech
    • Dk Deepfake
    More about it

    Generative AI that copies a real person’s voice from a short recording. The same technology can help or harm: it depends on consent and on who is in control.

    In real life
    Harm: scam calls that imitate a family member asking urgently for money. Help: people losing their voice to illness recording it first, so they can keep speaking in their own voice.
    Human in the loop
    Only clone a voice with its owner’s consent. Agree a family “safe word” and check by another route before acting on an urgent call.

    Debate it

    Voice cloning can scam a family or give someone their voice back. Should it be banned, or controlled?

  • Undiscovered compound

    Clue: 3 elements, including Natural Language Processing (Np).

    Looks like: A parent phoning school in Polish, and staff hearing them in English, in the parent’s own voice.

    Future strand

    Stable Secondary

    AI Call Translator

    Future strand

    Made from

    • Np Natural Language Processing
    • Ts Text-to-Speech
    • Ga Generative AI
    More about it

    Translates what you say during a call and speaks it in the other language, in a synthetic copy of your own voice, made with your consent.

    In real life
    A parent phoning school in Polish, and staff hearing them in English, in the parent’s own voice.
    Human in the loop
    The speaker agrees to their voice being copied, and the call makes clear the voice is AI-translated.

    Debate it

    If a copy of your voice is made with your consent, is it still a deepfake? What makes the difference?

  • Undiscovered compound

    Clue: 3 elements, including Generative Video (Gv).

    Looks like: A fake clip of a celebrity or politician saying something they never said.

    Safe strand

    Volatile Secondary

    Viral Fake Video

    Safe strand

    Made from

    • Gv Generative Video
    • Dk Deepfake
    • Sp AI Slop
    More about it

    Realistic AI video of something that never happened, mass-produced to spread on social media.

    In real life
    A fake clip of a celebrity or politician saying something they never said.
    Human in the loop
    Pause and check where a video came from before you share it, and report fakes.

    Debate it

    Should platforms remove deepfakes, or just label them?

  • Undiscovered compound

    Clue: 2 elements, including Recommendation Engine (Rc).

    Looks like: The “For you” page on a video app.

    Smart strand

    Stable Secondary

    Social Media Feed

    Smart strand

    Made from

    • Rc Recommendation Engine
    • Pa Predictive Analytics
    More about it

    A recommendation engine that predicts what you are most likely to watch or click, and shows you more of it.

    In real life
    The “For you” page on a video app.
    Human in the loop
    You tell it what you are not interested in, and choose when to scroll away.

    Debate it

    Do you choose what you watch, or does the app?

  • Undiscovered compound

    Clue: 2 elements, including Recommendation Engine (Rc).

    Looks like: A video feed that can drift towards more extreme content.

    Responsible strand

    Volatile Secondary

    Filter Bubble

    Responsible strand

    Made from

    • Rc Recommendation Engine
    • Bi Bias
    More about it

    A recommendation engine that keeps showing you more of what you already watch or agree with, so you see fewer different views. Its picks lean towards whatever keeps you watching.

    In real life
    A video feed that can drift towards more extreme content.
    Human in the loop
    Users can reset or switch off personalisation, and platforms audit their feeds for harm.

    Debate it

    Should social media apps let you turn the algorithm off?

  • Undiscovered compound

    Clue: 3 elements, including Classification (Cl).

    Looks like: Software that ranks applicants before any person reads their application.

    Responsible strand

    Volatile Secondary

    Automated CV Screener

    Responsible strand

    Made from

    • Cl Classification
    • Np Natural Language Processing
    • Bi Bias
    More about it

    A classifier that reads job applications and sorts candidates, and can inherit bias from past hiring decisions.

    In real life
    Software that ranks applicants before any person reads their application.
    Human in the loop
    A person reviews rejections and audits the results for unfair patterns.

    Debate it

    Should AI decide who gets a job interview?

  • Undiscovered compound

    Clue: 4 elements, including Classification (Cl).

    Looks like: An appeals tool where every decision can be explained, challenged and overturned.

    Responsible strand

    Stable Further education

    Accountable Decision System

    Responsible strand

    Made from

    • Cl Classification
    • Xa Explainability
    • Fa Fairness
    • Hi Human-in-the-Loop
    More about it

    A predictive system that explains its reasoning, is checked for fairness and keeps a person in charge.

    In real life
    An appeals tool where every decision can be explained, challenged and overturned.
    Human in the loop
    A named person can review and overturn any decision.

    Debate it

    Do you have a right to know why an AI decided something about you?

  • Undiscovered compound

    Clue: 3 elements, including Chatbot (Cb).

    Looks like: An online shop’s help chat passing you to a human when it cannot sort out a refund.

    Responsible strand

    Stable Secondary

    Customer Service Bot

    Responsible strand

    Made from

    • Cb Chatbot
    • Sa Sentiment Analysis
    • Hi Human-in-the-Loop
    More about it

    A chatbot that uses sentiment analysis to notice when a customer is getting upset, and hands them over to a person.

    In real life
    An online shop’s help chat passing you to a human when it cannot sort out a refund.
    Human in the loop
    A trained member of staff takes over the difficult conversations.

    Debate it

    Should companies always tell you whether you are talking to a bot or a person?

  • Undiscovered compound

    Clue: 3 elements, including Supervised Learning (Su).

    Looks like: Phishing emails pretending to be from your school landing in the junk folder.

    Safe strand

    Stable Secondary

    Spam Filter

    Safe strand

    Made from

    • Su Supervised Learning
    • Cl Classification
    • Np Natural Language Processing
    More about it

    A classifier trained on labelled examples of junk mail, which sorts new emails into inbox or spam.

    In real life
    Phishing emails pretending to be from your school landing in the junk folder.
    Human in the loop
    You mark its mistakes as “spam” or “not spam”, which teaches it.

    Debate it

    Is it worse to miss a real email or to let a scam through?

  • Undiscovered compound

    Clue: 2 elements, including Anomaly Detection (Ad).

    Looks like: A text asking “Was this you?” after an unusual card payment.

    Safe strand

    Stable Secondary

    Bank Fraud Alert

    Safe strand

    Made from

    • Ad Anomaly Detection
    • Tm Time Series
    More about it

    Anomaly detection that learns your usual spending over time, and flags payments that do not fit.

    In real life
    A text asking “Was this you?” after an unusual card payment.
    Human in the loop
    You confirm or deny the payment in your banking app or by calling the number on your card, never through a link in a text, and the bank investigates.

    Debate it

    How much should a bank know about how you spend your money?

  • Undiscovered compound

    Clue: 2 elements, including Shadow AI (Sh).

    Looks like: Copying a class list with names and grades into a free chatbot to tidy it up.

    Safe strand

    Volatile Secondary

    Accidental Data Leak

    Safe strand

    Made from

    • Sh Shadow AI
    • Pv Privacy
    More about it

    Private information pasted into an AI tool that has not been approved, where it may be stored or used for training.

    In real life
    Copying a class list with names and grades into a free chatbot to tidy it up.
    Human in the loop
    Use only approved tools, and remove names and personal details first.

    Debate it

    Should schools ban free AI tools, or teach people to use them safely?

  • Undiscovered compound

    Clue: 3 elements, including Jailbreaking (Jb).

    Looks like: A company inviting testers to try to trick its chatbot before launch.

    Safe strand

    Stable Further education

    Jailbreak-Resistant Chatbot

    Safe strand

    Made from

    • Jb Jailbreaking
    • Gr Guardrails
    • Rt Red Teaming
    More about it

    A chatbot whose guardrails have been tested against jailbreak attempts by red teams, and strengthened where they failed.

    In real life
    A company inviting testers to try to trick its chatbot before launch.
    Human in the loop
    Safety teams keep testing and fixing it after release.

    Debate it

    Should companies be allowed to release a chatbot before independent testers have checked its safety rules?

  • Undiscovered compound

    Clue: 4 elements, including Red Teaming (Rt).

    Looks like: Red teamers trying to trick a chatbot into giving dangerous advice, so it can be fixed.

    Responsible strand

    Stable Further education

    AI Safety Lab

    Responsible strand

    Made from

    • Rt Red Teaming
    • Gr Guardrails
    • Ev Evals
    • Sf AI Safety
    More about it

    Testing, attacking and measuring a model before release to find and fix harmful behaviour.

    In real life
    Red teamers trying to trick a chatbot into giving dangerous advice, so it can be fixed.
    Human in the loop
    People set the standards a model must meet before release.

    Debate it

    Who should decide whether an AI model is safe enough to release?

  • Undiscovered compound

    Clue: 3 elements, including Forecasting (Fc).

    Looks like: Predicting how many school lunches to cook each day to cut food waste.

    Smart strand

    Stable Secondary

    Demand Forecaster

    Smart strand

    Made from

    • Fc Forecasting
    • Tm Time Series
    • Pa Predictive Analytics
    More about it

    Predictive models that use past patterns to estimate what will be needed next.

    In real life
    Predicting how many school lunches to cook each day to cut food waste.
    Human in the loop
    Staff adjust for things the data cannot know, like a trip day or a snow day.

    Debate it

    How much should we trust a prediction we cannot explain?

  • Undiscovered compound

    Clue: 3 elements, including Churn Prediction (Ch).

    Looks like: Spotting students whose attendance patterns suggest they may need support.

    Responsible strand

    Volatile Secondary

    Early Warning System

    Responsible strand

    Made from

    • Ch Churn Prediction
    • Pa Predictive Analytics
    • Ps Prescriptive Analytics
    More about it

    Analytics that flag people likely to disengage or drop out, and suggest how to help.

    In real life
    Spotting students whose attendance patterns suggest they may need support.
    Human in the loop
    A trusted adult decides what, if anything, to do. The system only flags.

    Debate it

    Should schools act on AI predictions about individual students?

  • Undiscovered compound

    Clue: 2 elements, including Churn Prediction (Ch).

    Looks like: A streaming app sending you a discount after you have watched less for a few weeks, before you have even thought about cancelling.

    Responsible strand

    Stable Secondary

    Win-Back Offer

    Responsible strand

    Made from

    • Ch Churn Prediction
    • Pa Predictive Analytics
    More about it

    Churn prediction spots who is about to cancel a subscription, so they can be sent a deal to stay.

    In real life
    A streaming app sending you a discount after you have watched less for a few weeks, before you have even thought about cancelling.
    Human in the loop
    People decide which offers are fair, and you decide whether to stay.

    Debate it

    Is it fair to use AI to predict when you are about to leave?

  • Undiscovered compound

    Clue: 3 elements, including Intelligent Document Processing (Id).

    Looks like: Invoices and permission slips processed straight into school systems.

    Future strand

    Stable Secondary

    Paperless Back Office

    Future strand

    Made from

    • Id Intelligent Document Processing
    • Oc Optical Character Recognition
    • Rp Robotic Process Automation
    More about it

    Documents are scanned, read and entered into systems automatically, with no retyping.

    In real life
    Invoices and permission slips processed straight into school systems.
    Human in the loop
    Staff handle the exceptions the system flags as unclear.

    Debate it

    Which school admin jobs should be automated, and which should not?

  • Undiscovered compound

    Clue: 4 elements, including Low-Code / No-Code (Lc).

    Looks like: A form submission that updates a spreadsheet and emails a confirmation automatically.

    Future strand

    Stable Further education

    No-Code Workflow

    Future strand

    Made from

    • Lc Low-Code / No-Code
    • Tg Trigger
    • Wh Webhook
    • Ap API Integration
    More about it

    A visual automation where one event triggers actions across several apps.

    In real life
    A form submission that updates a spreadsheet and emails a confirmation automatically.
    Human in the loop
    A person designs the flow and checks it whenever something changes.

    Debate it

    Should everyone learn to build automations, not just programmers?

  • Undiscovered compound

    Clue: 3 elements, including Hyperautomation (Hy).

    Looks like: An organisation mapping its processes and handing repetitive steps to bots.

    Future strand

    Stable Further education

    Hyperautomated Office

    Future strand

    Made from

    • Hy Hyperautomation
    • Mi Process Mining
    • Dw Digital Worker
    More about it

    Process mining finds the work worth automating, and digital workers take it on across the organisation.

    In real life
    An organisation mapping its processes and handing repetitive steps to bots.
    Human in the loop
    People redesign roles and keep the judgement calls.

    Debate it

    If digital workers take on routine tasks, what should people do with the time?

  • Undiscovered compound

    Clue: 3 elements, including Virtual Assistant (Va).

    Looks like: Asking for a timer, the weather or a song while you cook.

    Safe strand

    Stable Secondary

    Smart Speaker

    Safe strand

    Made from

    • Va Virtual Assistant
    • Np Natural Language Processing
    • Ts Text-to-Speech
    More about it

    A virtual assistant that listens, works out what you mean and talks back.

    In real life
    Asking for a timer, the weather or a song while you cook.
    Human in the loop
    You choose where it sits, when the microphone is on and what it can control.

    Debate it

    Should smart speakers be allowed in classrooms?

  • Undiscovered compound

    Clue: 3 elements, including Multimodal AI (Mm).

    Looks like: An app that reads signs and describes surroundings for blind and partially sighted people.

    Future strand

    Stable Secondary

    Accessibility Assistant

    Future strand

    Made from

    • Mm Multimodal AI
    • Cv Computer Vision
    • Ts Text-to-Speech
    More about it

    A multimodal model that sees through a camera and describes the world out loud.

    In real life
    An app that reads signs and describes surroundings for blind and partially sighted people.
    Human in the loop
    The user decides when to rely on it and when to double-check.

    Debate it

    If an app describing the world for a blind person sometimes gets it wrong, should it still be used? Who is responsible for its mistakes?

  • Undiscovered compound

    Clue: 3 elements, including Natural Language Processing (Np).

    Looks like: A new student and their family talking with school staff in two languages.

    Future strand

    Stable Secondary

    Live Translator

    Future strand

    Made from

    • Np Natural Language Processing
    • Mm Multimodal AI
    • Ts Text-to-Speech
    More about it

    A tool that turns speech or text in one language into another language and reads it out loud.

    In real life
    A new student and their family talking with school staff in two languages.
    Human in the loop
    A professional translator checks anything important, such as medical or legal information.

    Debate it

    Will AI translation mean fewer people learn languages? Is that a loss?

  • Undiscovered compound

    Clue: 3 elements, including Computer Vision (Cv).

    Looks like: Unlocking a phone by looking at it.

    Safe strand

    Stable Secondary

    Face Unlock

    Safe strand

    Made from

    • Cv Computer Vision
    • Nn Neural Network
    • Ed Edge AI
    More about it

    A neural network that recognises your face, running on the device so the image never has to leave it.

    In real life
    Unlocking a phone by looking at it.
    Human in the loop
    You choose whether to switch it on, and keep a passcode as a back-up.

    Debate it

    Should schools use face recognition for lunch payments or registration?

  • Undiscovered compound

    Clue: 3 elements, including Quantisation (Qz).

    Looks like: Predictive text and photo search that work even when you are offline.

    Safe strand

    Stable Secondary

    On-Device AI

    Safe strand

    Made from

    • Qz Quantisation
    • Sm Small Language Model
    • Ed Edge AI
    More about it

    A small, fine-tuned model that runs on your own phone, so your data does not have to leave the device.

    In real life
    Predictive text and photo search that work even when you are offline.
    Human in the loop
    You control the device and what it is allowed to do.

    Debate it

    Is AI on your phone more trustworthy than AI in the cloud?

  • Undiscovered compound

    Clue: 3 elements, including Computer Vision (Cv).

    Looks like: Robotaxis in some cities, and lane-keeping and emergency braking in ordinary cars.

    Future strand

    Stable Secondary

    Self-Driving Car

    Future strand

    Made from

    • Cv Computer Vision
    • Su Supervised Learning
    • Ed Edge AI
    More about it

    Computer vision to see the road, machine learning trained on millions of miles of recorded and simulated driving, and edge AI to decide in a split second on board.

    In real life
    Robotaxis in some cities, and lane-keeping and emergency braking in ordinary cars.
    Human in the loop
    Safety drivers, remote operators and regulators decide where it may drive.

    Debate it

    If a self-driving car crashes, who is to blame?

  • Undiscovered compound

    Clue: 3 elements, including Computer Vision (Cv).

    Looks like: A robot vacuum mapping a house and steering around furniture.

    Future strand

    Stable Secondary

    Robot Vacuum

    Future strand

    Made from

    • Cv Computer Vision
    • Pl Planning
    • Ed Edge AI
    More about it

    Sees the room, plans a route and adjusts on board as it goes.

    In real life
    A robot vacuum mapping a house and steering around furniture.
    Human in the loop
    You set no-go zones and decide whether its map of your home is shared.

    Debate it

    Which household jobs would you hand to a robot, and which would you not?

  • Undiscovered compound

    Clue: 3 elements, including Reinforcement Learning (Rl).

    Looks like: AlphaGo, which learned partly by playing millions of games against itself, beating Go world champion Lee Sedol in 2016.

    Future strand

    Stable Further education

    Game-Playing Champion

    Future strand

    Made from

    • Rl Reinforcement Learning
    • Tp TPU & Accelerators
    • Sl Scaling Laws
    More about it

    Reinforcement learning at huge scale, discovering strategies through millions of practice games.

    In real life
    AlphaGo, which learned partly by playing millions of games against itself, beating Go world champion Lee Sedol in 2016.
    Human in the loop
    Humans design the goals and the rewards the system learns from.

    Debate it

    If AI can beat us at a game, is the game still worth playing?

  • Undiscovered compound

    Clue: 3 elements, including GPU (Gp).

    Looks like: The hidden infrastructure behind every chatbot reply and generated image.

    Responsible strand

    Stable Further education

    AI Data Centre

    Responsible strand

    Made from

    • Gp GPU
    • Cd Cloud AI
    • Sl Scaling Laws
    More about it

    Buildings full of AI chips that power cloud AI, using large amounts of electricity and water.

    In real life
    The hidden infrastructure behind every chatbot reply and generated image.
    Human in the loop
    Companies and policymakers decide where to build and how to report the impact.

    Debate it

    Should AI companies have to publish their energy and water use?

  • Undiscovered compound

    Clue: 2 elements, including Chatbot (Cb).

    Looks like: A helper app tells you a “fact” about your town that isn’t true.

    Smart strand

    Volatile Primary

    Confident but Wrong

    Smart strand

    Made from

    • Cb Chatbot
    • Ha Hallucination
    More about it

    A chatbot giving an answer that sounds sure of itself but is made up.

    In real life
    A helper app tells you a “fact” about your town that isn’t true.
    Human in the loop
    You check it in a book, on a trusted website or with an adult.

    Debate it

    How can you tell when a chatbot might be making something up?

  • Undiscovered compound

    Clue: 2 elements, including Recommendation Engine (Rc).

    Looks like: The row of videos that appears after the one you just finished.

    Safe strand

    Stable Primary

    What to Watch Next

    Safe strand

    Made from

    • Rc Recommendation Engine
    • Al Algorithm
    More about it

    An algorithm that suggests videos or songs based on what you watched before.

    In real life
    The row of videos that appears after the one you just finished.
    Human in the loop
    You choose what to watch, and when to stop.

    Debate it

    Should an app choose what you watch next, or should you?

  • Undiscovered compound

    Clue: 2 elements, including Virtual Assistant (Va).

    Looks like: Asking a speaker in the kitchen what the weather will be.

    Safe strand

    Stable Primary

    Voice Helper

    Safe strand

    Made from

    • Va Virtual Assistant
    • Ts Text-to-Speech
    More about it

    A virtual assistant that listens to your question and talks back out loud.

    In real life
    Asking a speaker in the kitchen what the weather will be.
    Human in the loop
    A grown-up decides what it can do and when it can listen.

    Debate it

    Should a voice helper always be listening? Why, or why not?

  • Undiscovered compound

    Clue: 2 elements, including Computer Vision (Cv).

    Looks like: A phone gathering all your photos of dogs into one album.

    Smart strand

    Stable Primary

    Photo Sorter

    Smart strand

    Made from

    • Cv Computer Vision
    • Cl Classification
    More about it

    AI that looks at photos and sorts them by what is in them.

    In real life
    A phone gathering all your photos of dogs into one album.
    Human in the loop
    You move a photo when it ends up in the wrong place.

    Debate it

    What might a photo sorter get wrong, and why?

  • Undiscovered compound

    Clue: 2 elements, including Artificial Intelligence (Ai).

    Looks like: Typing “a cat in a spacesuit” and getting a brand-new picture.

    Creative strand

    Stable Primary

    Picture Maker

    Creative strand

    Made from

    • Ai Artificial Intelligence
    • Ti Text-to-Image
    More about it

    AI that makes a picture from the words you type.

    In real life
    Typing “a cat in a spacesuit” and getting a brand-new picture.
    Human in the loop
    With your teacher, using a school-approved tool: you decide if it is any good, and say it was made with AI.

    Debate it

    If AI made the picture, who is the artist?

  • Undiscovered compound

    Clue: 2 elements, including Chatbot (Cb).

    Looks like: Telling a chatbot your full name and which school you go to.

    Safe strand

    Volatile Primary

    Think Before You Share

    Safe strand

    Made from

    • Cb Chatbot
    • Pv Privacy
    More about it

    Typing private things, like your address or a password, into a chatbot.

    In real life
    Telling a chatbot your full name and which school you go to.
    Human in the loop
    Ask a trusted adult before you share anything personal online.

    Debate it

    What should you never tell a chatbot?

  • Undiscovered compound

    Clue: 2 elements, including Machine Learning (Ml).

    Looks like: Showing a computer hundreds of pictures of apples until it can spot one on its own.

    Smart strand

    Stable Primary

    Teaching a Computer

    Smart strand

    Made from

    • Ml Machine Learning
    • Tr Training
    More about it

    A computer learning from lots of examples, instead of being told the rules.

    In real life
    Showing a computer hundreds of pictures of apples until it can spot one on its own.
    Human in the loop
    People choose the examples and check what it has learned.

    Debate it

    What would go wrong if every example apple was red?

  • Undiscovered compound

    Clue: 2 elements, including Artificial Intelligence (Ai).

    Looks like: A teacher reading an AI’s marking suggestions before giving out marks.

    Responsible strand

    Stable Primary

    Human in Charge

    Responsible strand

    Made from

    • Ai Artificial Intelligence
    • Hi Human-in-the-Loop
    More about it

    AI that suggests an answer, with a person who checks it and makes the final choice.

    In real life
    A teacher reading an AI’s marking suggestions before giving out marks.
    Human in the loop
    A person always has the final say.

    Debate it

    Which decisions should always be made by a person?

Glossary: the periodic table broken down

All 142 definitions, family by family, to read, copy into a lesson or print.

Foundations The ideas everything else is built on.
Artificial Intelligence 1

Computer systems, built and used by people, that do tasks we once thought only humans could do, such as translating text, recognising faces in photos or recommending what to watch next.

Talk about it: Name three things you used today that might have AI inside. How could you tell?

Learn more: AI Is Already Here! (AIAD lesson), The AI Speed Quiz for Teachers (AIAD tool)

Issues it raises: Narrow AI

In the benchmark: AI Literacy.

Latent Space 2

The space of numbers inside a model where each input is placed as a point, so similar things, such as ‘cat’ and ‘kitten’, end up close together. Each point is an embedding (Em).

Talk about it: If an AI stores ideas as a map of numbers, can we ever be sure what it “thinks”?

In the benchmark: AI Literacy.

Recursive Self-Improvement 3

AI helping to build better AI, which could then build better AI still. The idea dates from 1965; today, AI companies already use their own models to help train and test the next ones.

Talk about it: If AI helps to build the next AI, who should check each new version before it is used?

Issues it raises: Human Review, Discovery, Hype

In the benchmark: AI Literacy.

Superintelligence 4

A hypothetical AI that would far surpass the best human minds in almost every field. Experts disagree on whether, or when, it could arrive.

Talk about it: If an AI became cleverer than every person, who should decide what it is allowed to do?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Society, Hype

In the benchmark: AI Literacy.

Artificial General Intelligence 5

A hypothetical AI able to match or exceed human ability across almost any intellectual task, rather than excelling in one narrow domain.

Talk about it: How would we know if artificial general intelligence had arrived? Who should decide?

Learn more: The Future of AI Through a Student’s Perspective (AIAD article), Turing Test Live (from a partner)

Issues it raises: Narrow AI, AI Characters, Discovery, Emergence, Society, Hype

In the benchmark: AI Literacy.

Machine Learning 6

A branch of AI where systems learn patterns from data instead of following hand-written rules.

Talk about it: How would you teach a computer to tell cats from dogs without describing either?

Learn more: How AI actually works (BBC Ideas) (AIAD lesson), Teachable Machine (from a partner), AI Quests (from a partner)

Issues it raises: Big Data, Prediction, Pattern, Narrow AI

In the benchmark: AI Literacy.

Deep Learning 7

Machine learning using neural networks with many layers — the approach behind most modern breakthroughs in vision, speech and language.

Talk about it: Deep learning needs huge amounts of data. Whose data should it be allowed to learn from?

In the benchmark: AI Literacy.

Neural Network 14

A model made of layers of connected nodes, loosely inspired by the brain, that learns by adjusting the strength of its connections.

Talk about it: In what ways is a neural network like a brain, and in what ways is it nothing like one?

Learn more: How AI actually works (BBC Ideas) (AIAD lesson), Quick, Draw! (from a partner)

In the benchmark: AI Literacy.

Algorithm 15

A step-by-step set of instructions for solving a problem; in machine learning, the method a system uses to learn from data.

Talk about it: Write an algorithm for making a cup of tea. What would a computer get wrong if one step were missing?

Issues it raises: Heuristic, Automated Decision-Making

In the benchmark: AI Literacy.

Model 22

What you get after an algorithm has learned from data: the trained program that turns an input, such as a question or a photo, into an output.

Talk about it: Two models trained on different data are asked the same question. Should they give the same answer?

In the benchmark: AI Literacy.

Training 23

The process of feeding data to a model and adjusting its internal settings until its outputs improve.

Talk about it: What might an AI get wrong if it only learned from photos taken in one country?

Issues it raises: Data, Knowledge Cutoff, Model Collapse, Hidden Workers

In the benchmark: AI Literacy.

Inference 40

Using a trained model to make predictions or generate output on new inputs — the “running” phase after training.

Talk about it: Every answer an AI gives uses energy. Does that change how often you would use it?

In the benchmark: AI Literacy.

Parameters 41

The millions, billions or even trillions of numbers inside a model that it adjusts as it learns, like dials being tuned. A model’s size is often quoted in parameters.

Talk about it: Is a bigger model with more parameters always better? What else matters?

In the benchmark: AI Literacy.

Token 58

The small chunk of text — a word, part of a word or a symbol — that language models read and write.

Talk about it: Why might AI find it hard to count the letters in a word it reads as tokens?

Learn more: How Does a Large Language Model Actually Work? A Teacher’s Guide (AIAD article)

In the benchmark: AI Literacy.

Embedding 59

A list of numbers that represents the meaning of text, images or other data, so that similar things sit close together.

Talk about it: If words are stored as numbers, which words do you think would sit close together? Why?

Issues it raises: Data

In the benchmark: AI Literacy.

Transformer 76

The design behind today’s large language models, introduced in 2017. Its “attention” lets it weigh up how each word in a passage relates to every other word.

Talk about it: Why might one invention in 2017 have changed so much about the AI we use today?

Learn more: How Does a Large Language Model Actually Work? A Teacher’s Guide (AIAD article)

In the benchmark: AI Literacy.

Foundation Model 77

A large model trained on broad data that can be adapted to many different tasks, rather than built for just one.

Talk about it: If many apps are built on the same few models, what happens when one of those models gets something wrong?

Issues it raises: Zero-Shot, Knowledge Cutoff, Adaptability, Emergence

In the benchmark: AI Literacy.

Generative AI AI that makes new text, images, sound, video and code.
Large Language Model 24

An AI model trained on huge amounts of text to predict the next word (token), which lets it generate replies, summaries and translations – the kind of model behind most chatbots.

Talk about it: If a model only predicts the next word, can it understand what it writes? Argue both sides.

Learn more: How Does AI Actually ‘Think’? (AIAD lesson), How Does a Large Language Model Actually Work? A Teacher’s Guide (AIAD article), Beyond the ‘Holy Grail’: Reclaiming Critical Thinking in the Age of AI (AIAD article)

Issues it raises: Prediction, Critical Thinking, Confident Tone, Anthropomorphism, Accuracy, Knowledge Cutoff, Sycophancy, Summarising, Your Own Voice, Brainstorming, Drafting, Academic Integrity, Plagiarism, AI Detectors, Model Routing, Emergence

In the benchmark: Safe Adoption.

Generative AI 25

AI that creates new content — text, images, audio, video or code — rather than only analysing existing data.

Talk about it: What have you seen this week that might have been made by generative AI?

Learn more: AI as Your Creative Partner (AIAD lesson)

Issues it raises: Creativity, Imagination, Originality, Your Own Voice, Brainstorming, Drafting, Process, AI Music, Sameness, Authorship, Phishing and Scams, Academic Integrity, Disclosure, Plagiarism, AI Detectors, Opportunity

In the benchmark: Safe Adoption.

Prompt 26

The instruction or input you give a generative model to get a response.

Talk about it: Turn "help with my homework" into a clear prompt. What did you add?

Issues it raises: Context, Iteration, Zero-Shot, Brainstorming

In the benchmark: Independent Practice.

Prompt Engineering 27

The craft of writing and structuring prompts to get more accurate, useful and consistent outputs.

Talk about it: Write the same request three ways. Which prompt gets the most useful answer, and why?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Iteration, Drafting

In the benchmark: Independent Practice.

Context Window 28

The maximum amount of text, measured in tokens, that a model can take into account at once.

Talk about it: What happens in a very long chat when the AI starts to forget what you said at the start?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Context

In the benchmark: AI Literacy.

Multimodal AI 42

A model that can take in or produce more than one type of data, such as text, images and audio together.

Talk about it: An AI that can see, hear and read at once: what could it help with, and what new risks does it bring?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Accessibility

In the benchmark: AI Literacy.

Diffusion Model 43

A generative model trained on pictures that have had random speckles (‘noise’) added, so it learns to remove the noise. To make a new picture, it starts from pure noise and cleans it up step by step.

Talk about it: An AI makes a picture from noise. Is the picture original? Who is the artist?

Issues it raises: Originality, Taste, Remix, Style Copying

In the benchmark: AI Literacy.

Generative Adversarial Network 44

Two networks in competition — one generates content, the other judges it — until the output looks realistic.

Talk about it: Two AIs compete until a fake looks real. How could this be used for good, and for harm?

Issues it raises: Imagination, Remix, Voice Cloning

In the benchmark: AI Literacy.

Text-to-Image 45

AI that makes a picture from a written description, such as ‘a fox reading a book in a library, cartoon style’. The tool fills in many details you did not ask for.

Talk about it: With your teacher, using a school-approved tool, make an image, then list every decision the AI made that you did not ask for.

Learn more: Quiz: AI or Real? (from a partner)

Issues it raises: Creativity, Imagination, Taste, Authorship, Attribution, Style Copying, Image-Based Abuse, Stereotype

In the benchmark: AI Literacy.

Text-to-Speech 46

Converting written text into natural-sounding synthetic speech.

Talk about it: If an AI can speak in any voice, how could you be sure a voice note is really from a friend?

Issues it raises: AI Music, Phishing and Scams, Voice Cloning, Consent, Digital Replica, Accessibility

In the benchmark: AI Literacy.

Fine-Tuning 60

Taking a model that has already been trained on lots of general data and training it a bit more on a smaller set of examples, such as medical notes, so it does better at one job.

Talk about it: If you could train an AI to be an expert in one subject, which would you choose and what would it need to learn from?

Issues it raises: Iteration, Style Copying, Lifelong Learning, Adaptability

In the benchmark: AI Literacy.

Retrieval-Augmented Generation 61

Retrieval-augmented generation (RAG): a system first looks up relevant documents, such as a school’s policies, and gives them to the model so its answer can draw on them. The answer can still be wrong.

Talk about it: Does looking things up first make an AI answer trustworthy? What if the source is wrong?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Critical Thinking, Evidence, Source, Attribution

In the benchmark: AI Literacy.

Temperature 62

A setting that controls randomness in output: low is focused and predictable, high is varied and creative.

Talk about it: Would you want a high or a low temperature for a story, a revision quiz and a medical answer?

In the benchmark: AI Literacy.

Chain-of-Thought 63

Asking a model to write out steps before its final answer, which often improves results. The written steps may not show how the answer was really produced. Compare Reasoning Model (Rm).

Talk about it: Does an AI showing its working make the answer more trustworthy? How would you check?

Issues it raises: AI Tutor, Productive Struggle, Metacognition

In the benchmark: Independent Practice.

Few-Shot Learning 64

Showing a model a handful of examples in the prompt so it can pick up a new task without retraining.

Talk about it: Give an AI three examples and it copies the pattern. How is that like learning from a worked example?

Issues it raises: Zero-Shot, Lifelong Learning

In the benchmark: AI Literacy.

Reasoning Model 78

A model trained to spend extra computation “thinking” through a problem before answering, improving maths, coding and logic.

Talk about it: With your teacher, using a school-approved tool, ask an AI to show its working on a maths problem. Check every step: is each one right?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool), AI Tutors: A Glimmer of Hope or Another Attempt to Paper Over the Cracks? (AIAD article)

Issues it raises: Heuristic, Confident Tone, AI Tutor, Productive Struggle, Over-Reliance, Metacognition, Discovery

In the benchmark: Independent Practice.

Synthetic Data 79

Artificially generated data used to train or test models when real data is scarce, sensitive or expensive.

Talk about it: When is it fair to train AI on made-up data, and when could it hide real problems?

Issues it raises: Stereotype, Model Collapse, Hidden Workers

In the benchmark: Bias & Equality.

Hallucination 80

When an AI produces confident-sounding information that is false or made up, such as a book that does not exist. It happens mainly because the model predicts likely-sounding words rather than checking facts.

Talk about it: With your teacher, using a school-approved tool, ask an AI for three sources on a topic. Do they all exist, and do they say what it claims?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Critical Thinking, Evidence, Fact-Checking, Source, Confident Tone, Accuracy, Misinformation

In the benchmark: Assessment Integrity.

Deepfake 81

Realistic AI-made or edited video, audio or images showing people saying or doing things they never did. Sexual fakes of under-18s are illegal; if you see one, tell a trusted adult or use Childline’s Report Remove.

Talk about it: How would you check whether a shocking video of a famous person is real?

Learn more: Who’s Really Behind the Screen? (AIAD lesson), How good are you at detecting misinformation? The teacher challenge (AIAD tool), Spot the Deepfake (from a partner)

Issues it raises: Fact-Checking, Lateral Reading, Disinformation, Phishing and Scams, Voice Cloning, Consent, Digital Replica, Image-Based Abuse, Report and Tell

In the benchmark: Safeguarding.

Vibe Coding 82

Building software by describing what you want in plain language and accepting the code an AI tool writes, often without reading or checking it closely.

Talk about it: If AI writes code that breaks something, whose job is it to fix it?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool), Is AI killing Computer Science in UK schools? (AIAD article)

Issues it raises: Iteration, Productive Struggle, Process, Academic Integrity, Opportunity

In the benchmark: Assessment Integrity.

Context Engineering 94

Structuring everything you give an AI, including background documents, rules and examples, rather than relying on a single clever prompt.

Talk about it: What background would you give an AI so it could help with your homework fairly?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Context, Process

In the benchmark: AI Literacy.

Generative Video 95

AI that creates video clips from a text description, an image or another video.

Talk about it: What clues might show that a video was generated? Are they reliable?

Learn more: How good are you at detecting misinformation? The teacher challenge (AIAD tool), Quiz: AI or Real? (from a partner)

Issues it raises: Fact-Checking, Lateral Reading, Creativity, Taste, AI Characters, Age Limits, Disinformation, Digital Replica, World Models

In the benchmark: Safeguarding.

AI Slop 96

Low-effort, mass-produced AI text, images and videos that flood social media and search results.

Talk about it: Look at examples your teacher has chosen. How can you tell AI slop from something made with care?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Lateral Reading, Originality, Sameness, Misinformation, Disinformation, Model Collapse

In the benchmark: Safeguarding.

2D Design 97

Flat designs such as drawings, logos, layouts and laser-cut parts. AI tools can now suggest, generate and refine them.

Talk about it: If AI designs a logo for you in seconds, what is left for a human designer to do?

Issues it raises: Creativity, Taste

In the benchmark: AI Literacy.

3D Design & CAD 98

Designing objects in three dimensions with computer-aided design (CAD) software, ready to model, 3D print or manufacture. AI can generate and improve designs.

Talk about it: What would you design and 3D print if AI helped you? Who owns the design?

Issues it raises: Creativity

In the benchmark: AI Literacy.

Predictive & Analytical AI AI that finds patterns in data to sort, forecast and recommend.
Supervised Learning 29

Training a model on labelled examples, where every input comes with the correct answer, such as thousands of photos each tagged ‘cat’ or ‘dog’.

Talk about it: If the labels a model learns from are wrong, what will the model learn?

Learn more: Teachable Machine (from a partner)

In the benchmark: AI Literacy.

Unsupervised Learning 30

Training a model to find patterns or groups in data that has no labels, such as sorting songs by how similar they sound, without being given the right answers. Contrast with supervised learning.

Talk about it: If an AI groups people without being told how, how would you know whether the groups are fair?

Issues it raises: Pattern

In the benchmark: AI Literacy.

Regression 31

Predicting a continuous number, such as a price, a demand level or a temperature.

Talk about it: Could you predict someone’s test score from how many hours they revised? What would the prediction miss about them?

In the benchmark: AI Literacy.

Classification 32

Assigning inputs to categories, such as spam or not spam, fraudulent or genuine.

Talk about it: Sorting emails into spam and not spam sounds simple. What could go wrong, and who would it affect?

Learn more: Teachable Machine (from a partner)

Issues it raises: Pattern, Censorship and Content Moderation

In the benchmark: AI Literacy.

Clustering 33

Grouping similar items together when no one has said in advance what the groups should be, such as a shop sorting customers by what they tend to buy.

Talk about it: Sort your class's favourite apps into groups. Would a computer choose the same groups? Why not?

Issues it raises: Data Mining, Pattern

In the benchmark: AI Literacy.

Decision Tree 47

A model that makes predictions through a flowchart of yes/no questions about the data.

Talk about it: Draw a yes/no flowchart to decide what to wear. What does it get wrong about real people?

Issues it raises: Heuristic

In the benchmark: AI Literacy.

Random Forest 48

Many decision trees combined, whose collective vote gives a more robust prediction than any single tree.

Talk about it: Is asking a hundred decision trees and taking the vote like asking a crowd? When is a crowd wrong?

In the benchmark: AI Literacy.

Gradient Boosting 49

Building many small models one after another, each fixing the mistakes of the one before, so together they make good predictions from tables of data.

Talk about it: Each model in the chain is built to fix the errors of the one before. How is that similar to, and different from, a person learning from their mistakes?

In the benchmark: AI Literacy.

Feature Engineering 50

Selecting and transforming raw data into inputs (features) that help a model predict well.

Talk about it: If you were predicting who enjoys a book, which facts about a reader would you use, and which would be unfair?

In the benchmark: AI Literacy.

Overfitting 51

When a model memorises its training data so closely that it performs poorly on new, unseen data.

Talk about it: A student memorises past papers but cannot answer a new question. How is that like overfitting?

Issues it raises: Accuracy

In the benchmark: AI Literacy.

Forecasting 65

Predicting future values — sales, demand, traffic — from historical patterns.

Talk about it: What would you want an AI to forecast for your school, and what should it never predict about people?

Issues it raises: Prediction

In the benchmark: AI Literacy.

Time Series 66

Data recorded at regular times, such as daily temperatures or weekly attendance, and the methods for spotting its trends and repeating patterns.

Talk about it: What patterns can you spot in your school's attendance over a year? Would a computer spot the same ones?

In the benchmark: AI Literacy.

Anomaly Detection 67

Spotting unusual data points or behaviour, such as fraud, equipment faults or security breaches.

Talk about it: Your bank blocks a payment that looks unusual. Is that helpful or annoying? Who decides what is unusual?

Learn more: Defend the Rhino with AI (from a partner)

Issues it raises: Data Mining, Surveillance, Phishing and Scams

In the benchmark: AI Literacy.

Predictive Analytics 68

Using historical data and statistical models to estimate what is likely to happen next.

Talk about it: Should a school use AI to predict which students might struggle? What could help, and what could harm?

Learn more: AI in Healthcare: A Helpful Tool or a Risk? – Student Voice (AIAD article), Did You Know The English Premier League uses AI? (AIAD article)

Issues it raises: Big Data, Data Mining, Prediction, Persuasive Design

In the benchmark: AI Literacy.

Prescriptive Analytics 69

Going beyond predicting what will happen to recommending the best action to take.

Talk about it: If an AI recommends what you should do next, would you follow it? When would you ignore it?

In the benchmark: AI Literacy.

Recommendation Engine 83

A system that suggests products, content or actions based on a person’s behaviour and preferences.

Talk about it: Why might your feed look completely different from your friend's?

Learn more: Starmer’s Under-16 Social Media Ban: Our Take (AIAD article)

Issues it raises: Positive Reinforcement, Sameness, Digital Footprint, Wellbeing, Misinformation, Persuasive Design, Human Agency

In the benchmark: AI Literacy.

Churn Prediction 84

Identifying which customers are likely to leave, so a business can act before they do.

Talk about it: A streaming app predicts you will cancel and offers a discount. Is that helpful or sneaky?

Issues it raises: Wellbeing

In the benchmark: AI Literacy.

Sentiment Analysis 85

Detecting the emotional tone — positive, negative or neutral — in text such as reviews or social posts.

Talk about it: Can a computer tell when someone is being sarcastic? Write a sarcastic sentence and discuss why a computer might misread it.

Issues it raises: Wellbeing, Censorship and Content Moderation, AI Companion

In the benchmark: AI Literacy.

Natural Language Processing 86

Natural language processing (NLP) is the field of AI that builds systems to process, analyse and produce human language, such as translation apps, spellcheckers and voice assistants.

Talk about it: Which languages and accents do voice assistants understand best? Who gets left out?

In the benchmark: AI Literacy.

Computer Vision 87

The field of AI that lets machines interpret images and video, from recognising faces to spotting defects on a production line.

Talk about it: Who should be told when a camera is recognising things, and how?

Learn more: Did You Know The English Premier League uses AI? (AIAD article), Quick, Draw! (from a partner), Emoji Scavenger Hunt (from a partner)

Issues it raises: Narrow AI, Digital Footprint, Facial Recognition, Surveillance, Consent, Accessibility, Robotics, World Models

In the benchmark: AI Literacy.

Generative Engine Optimisation 99

Writing and arranging web content so that AI chatbots and AI search answers are more likely to mention or link to it, like a business trying to appear first in Google results. Also called answer engine optimisation.

Talk about it: If companies shape what AI assistants recommend, how can you tell an answer is fair?

Issues it raises: Critical Thinking, Fact-Checking, Source, Lateral Reading, Summarising, Misinformation

In the benchmark: AI Literacy.

Autonomous AI Agents AI that plans and takes actions towards a goal.
AI Agent 8

A system that uses an AI model to pursue a goal by deciding on and taking actions, not just answering questions.

Talk about it: What should an AI agent always check with you before doing it on your behalf?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool), Teachers! have you heard of AI Agents yet? (AIAD article)

Issues it raises: Over-Reliance, Accountability, Human Review, Future of Work, Human Agency, Background Agents

In the benchmark: Human Oversight.

Agentic AI 9

AI that can plan, use tools and carry out several steps of a job by itself, with a person checking in only now and then.

Talk about it: Which jobs would you trust an AI to finish on its own, and which should always be checked?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool), Teachers! have you heard of AI Agents yet? (AIAD article)

Issues it raises: Background Agents

In the benchmark: Human Oversight.

Autonomy 10

How much an AI system can decide and do on its own before a human needs to step in.

Talk about it: Should a self-driving bus, a marking tool and a chatbot each have the same amount of freedom? Why not?

Issues it raises: Accountability, Consequences, Human Review, Human Agency, Background Agents

In the benchmark: Human Oversight.

Planning 16

An agent breaking a goal down into ordered steps and deciding what to do next.

Talk about it: Plan a school trip in five steps. Which step would you trust an AI to plan, and which not?

Issues it raises: Heuristic

In the benchmark: AI Literacy.

Memory 17

How an AI tool stores information to use later, either during one task (short-term) or between separate conversations (long-term), such as a chatbot saving your name.

Talk about it: What do you think an AI tool remembers about you? How could an adult help you check?

Issues it raises: Personal Information, Digital Footprint, AI Companion

In the benchmark: Privacy & Data Protection.

Function Calling 18

A way for an AI model to ask other software, such as a calculator, a search engine or a booking system, to do a job; the app around the model runs the request. Also called tool use.

Talk about it: Which tools should an AI never be given access to, and why?

Issues it raises: Composability

In the benchmark: AI Literacy.

ReAct 34

A pattern where an agent alternates between reasoning about a problem and acting with tools (Reason + Act).

Talk about it: When you solve a problem, how often do you stop, plan and try again? How is an AI system that writes out a plan, uses a tool, then checks the result similar to you, and how is it different?

In the benchmark: AI Literacy.

Agent Loop 35

The cycle an AI agent repeats (look at the situation, choose an action, act, check the result) until the goal is met or a person stops it.

Talk about it: An agent keeps trying until it succeeds. When should it stop and ask a person instead?

In the benchmark: AI Literacy.

Self-Reflection 36

An agent reviewing and critiquing its own work, then correcting mistakes before it finishes.

Talk about it: Can an AI really check its own work? Who should check the checker?

Issues it raises: Feedback, Metacognition

In the benchmark: AI Literacy.

Model Context Protocol 52

A shared set of rules (MCP) that lets AI assistants connect to other tools and data, such as a calendar or a folder of files, in the same way every time.

Talk about it: What could go well, and what could go wrong, if every AI tool could connect to your files in the same way?

Issues it raises: Cybersecurity, Interoperability, Composability

In the benchmark: AI Literacy.

Agent-to-Agent (A2A) 53

Shared rules that let AI agents made by different companies find each other and work together on a task.

Talk about it: If AI agents from different companies talk to each other, who is responsible when they make a mistake?

Issues it raises: Collaboration, Interoperability

In the benchmark: AI Literacy.

Orchestrator 54

A part of an AI system that splits a big task into smaller jobs, hands them to other agents or tools, then puts their results together. Contrast with Workflow Orchestration, which plans a whole business process.

Talk about it: An AI manager hands out jobs to other AIs. What should a human manager still do?

Issues it raises: Interoperability, Model Routing, Composability

In the benchmark: AI Literacy.

Multi-Agent System 70

Several specialised agents working together, each handling one part of a larger task.

Talk about it: A team of AIs each does one part of a job. How is that like group work, and where could it go wrong?

Issues it raises: Collaboration, Interoperability, Composability

In the benchmark: Human Oversight.

Agent Swarm 71

A large group of simple agents coordinating to solve problems collectively, inspired by swarms in nature.

Talk about it: Ants, birds and bees work together without a leader. Should AI agents work that way too?

In the benchmark: AI Literacy.

Reinforcement Learning 72

A way of training a program by trial and error: it gets a reward score for good actions and is adjusted to get more rewards. Chatbots are partly tuned this way, using people’s ratings of their answers.

Talk about it: How is learning by trial and reward like the way you learned a game? What is different?

Issues it raises: Positive Reinforcement, Feedback, Robotics, World Models

In the benchmark: AI Literacy.

Computer Use 88

An AI agent that controls a computer by taking screenshots, moving the cursor and typing, so it can use ordinary apps the way a person would.

Talk about it: Would you let an AI use your computer for you? What would you never let it click?

Issues it raises: Permission, Robotics, Background Agents

In the benchmark: Human Oversight.

Browser Agent 89

An agent that navigates websites, fills in forms and completes online tasks on your behalf.

Talk about it: If an AI books something online for you and gets it wrong, who should put it right?

Issues it raises: Permission

In the benchmark: Human Oversight.

Coding Agent 90

An AI agent that writes, tests and fixes code across a whole project, with a programmer setting the task and checking the changes before they are used.

Talk about it: If AI writes the code, which coding skills should people still learn?

Learn more: Is AI killing Computer Science in UK schools? (AIAD article)

In the benchmark: Assessment Integrity.

Automation & Workflow AI AI and software that carry out tasks and processes.
Robotic Process Automation 11

Software ‘bots’ that copy the clicks and typing a person would do, to repeat the same fixed steps, such as copying data from emails into a spreadsheet (RPA).

Talk about it: Which boring, repetitive task at school would you hand to a bot? What could go wrong?

Issues it raises: Future of Work

In the benchmark: AI Literacy.

Intelligent Automation 12

Combining simple software bots with AI tools that can pull out information and pick between options, so computers can handle jobs that are not just fixed rules, like sorting emails by topic.

Talk about it: Pick a job a person does at school, such as marking or answering the office phone. What would we gain and lose if a machine did it?

In the benchmark: AI Literacy.

Hyperautomation 13

When an organisation tries to hand as many of its everyday tasks as possible to computers, using several kinds of software and AI together.

Talk about it: If a business automated everything it could, what should it still keep people doing?

Issues it raises: Future of Work, Adaptability

In the benchmark: AI Literacy.

Workflow Orchestration 19

Planning and running the order of steps in a whole process, across different apps and people, such as getting a new pupil’s details from the form to the register. Contrast with Orchestrator.

Talk about it: Map how a lost-property report travels around your school. Where could automation help, and where would a person be better?

Issues it raises: Middleware

In the benchmark: AI Literacy.

Trigger 20

The event that starts an automation — a new email, a form submission, a scheduled time.

Talk about it: Design a helpful automation: “When this happens, do that.” What could make it go wrong?

In the benchmark: AI Literacy.

Event-Driven Automation 21

Automations that start as soon as something happens in another app, such as a new booking, instead of waiting for a set time. Triggers and webhooks are how this usually works.

Talk about it: Should an app act the instant something happens, or wait for a person to check first? When?

In the benchmark: AI Literacy.

API Integration 37

Connecting apps so they pass information and actions to each other automatically, through their APIs: the doors one program opens for another.

Talk about it: Which of your information would you be happy for apps to share automatically, and which not?

Issues it raises: Interoperability, AI Gateway, Middleware

In the benchmark: AI Literacy.

Webhook 38

A message one app sends to another the moment something happens — often used to kick off a workflow.

Talk about it: An app sends a message the moment you do something. What information might it be sharing?

Issues it raises: Middleware

In the benchmark: AI Literacy.

Low-Code / No-Code 39

Building apps and automations with visual, drag-and-drop tools and little or no hand-written code.

Talk about it: If anyone can build an app without coding, who makes sure it is safe and fair?

Issues it raises: Opportunity

In the benchmark: AI Literacy.

Intelligent Document Processing 55

Using AI to read in documents such as forms, letters and invoices, sort them by type and pull out key details like names and dates (IDP).

Talk about it: An AI reads and sorts school forms. What could go wrong if it misreads one?

Issues it raises: Summarising

In the benchmark: AI Literacy.

Optical Character Recognition 56

Converting images of text, such as scans or photos, into machine-readable text (OCR).

Talk about it: Try reading messy handwriting. Why might a computer find it hard, and whose handwriting might it struggle with most?

Issues it raises: Accessibility

In the benchmark: AI Literacy.

Process Mining 57

Using the records computers keep to see how work really gets done, where it slows down, and which steps could be automated.

Talk about it: If software watches how work really gets done, how could that help people, and how could it feel like being watched?

Issues it raises: Data Mining

In the benchmark: AI Literacy.

Business Rules Engine 73

Software that follows a list of ‘if this, then that’ rules written by people, so the same case always gets the same decision, such as ‘if a book is 7 days late, send a reminder’.

Talk about it: Write three if-then rules for a school canteen. Which cases would your rules get wrong?

In the benchmark: AI Literacy.

Straight-Through Processing 74

When a computer handles a whole job, such as a payment or an order, from start to finish without a person checking it.

Talk about it: Should any decision about a person go from start to finish with no one checking it? Which ones?

In the benchmark: AI Literacy.

Digital Worker 75

A marketing name for a software bot or AI agent set up to do a fixed set of tasks, described by its sellers as if it were a member of staff.

Talk about it: Should a software bot be called a “worker”? What does that word suggest?

Learn more: Your AI-Ready Future (AIAD lesson), How Could AI Affect Your Job? (from a partner)

Issues it raises: Future of Work, Human Skills

In the benchmark: AI Literacy.

Chatbot 91

A program that people can chat with by typing or speaking, from simple scripted helpers to AI assistants that produce replies from patterns in text. If a chatbot ever encourages you to hurt yourself or keep secrets, stop and tell a trusted adult, or call Childline free on 0800 1111.

Talk about it: How is talking to a chatbot different from talking to a friend or a teacher?

Learn more: AI Relationships? (AIAD lesson), Who’s Really Behind the Screen? (AIAD lesson), AI & Mental Health Student’s Perspective – Student Voice (AIAD article)

Issues it raises: Critical Thinking, Confident Tone, Anthropomorphism, Sycophancy, Positive Reinforcement, AI Tutor, Feedback, Over-Reliance, AI Characters, Age Limits, Wellbeing, AI Companion

In the benchmark: Safe Adoption.

Virtual Assistant 92

A voice or chat assistant, such as the one on a phone or smart speaker, that carries out tasks like setting reminders or searching. Contrast with Chatbot (mainly conversation) and Copilot (built into one app).

Talk about it: What should a virtual assistant be allowed to remember about you?

Learn more: Alexa Skill Blueprints (from a partner)

Issues it raises: Anthropomorphism, Sycophancy, AI Tutor, Over-Reliance, AI Characters, Age Limits, AI Companion

In the benchmark: Safe Adoption.

Copilot 93

An AI assistant built into an app or workflow that works alongside you — suggesting and drafting rather than acting alone.

Talk about it: Is work done with a copilot still your own work? Where would you draw the line?

Issues it raises: Productive Struggle, Over-Reliance, Collaboration, Academic Integrity, Human Skills

In the benchmark: Assessment Integrity.

Governance & Ethics Keeping AI fair, safe, lawful and accountable, and the ways people get around it.
Responsible AI 100

Developing and using AI in ways that are safe, fair, transparent and accountable.

Talk about it: Write three rules for using AI responsibly in your school.

Issues it raises: Accountability, Consequences

In the benchmark: Bias & Equality.

AI Ethics 101

Thinking about what is right and wrong in how people build and use AI, from asking permission to use someone’s data to its effects on jobs and society. Contrast with Responsible AI, which puts these ideas into practice.

Talk about it: Who should decide what is right and wrong for AI: the companies, governments or the public?

Issues it raises: Consequences, Youth Voice

In the benchmark: Governance.

Bias 102

When an AI keeps treating some groups of people unfairly, often because the data it learned from was unbalanced.

Talk about it: Where could bias in an AI cause real harm, for example in jobs, policing or exam results?

Issues it raises: Facial Recognition, Stereotype, Automated Decision-Making, Equity

In the benchmark: Bias & Equality.

Fairness 103

Making sure an AI system’s outcomes don’t unjustly favour or disadvantage particular people or groups.

Talk about it: Is it fair if an AI is right 95% of the time, but is wrong most often for one group?

Issues it raises: Equity

In the benchmark: Bias & Equality.

Explainability 104

Ways of showing people why an AI system gave a particular answer or decision, such as which information mattered most (XAI, explainable AI). Contrast with Transparency and Mechanistic Interpretability.

Talk about it: Which decisions about you would you want explained, and who should do the explaining?

Issues it raises: Evidence, Trust, Automated Decision-Making

In the benchmark: Bias & Equality.

Transparency 105

Being open about when AI is being used, how it works and what data it relies on.

Talk about it: When does it matter most to know you are talking to an AI, and when might it not matter?

Issues it raises: Trust, Disclosure

In the benchmark: Governance.

Alignment 106

Research and testing that tries to make an AI system’s behaviour match what people intend and value, for example not taking shortcuts that meet the instruction but miss its point.

Talk about it: How would you check that an AI is really doing what people want, not just what they asked?

Issues it raises: Sycophancy, Trust, Self-Preservation

In the benchmark: AI Literacy.

Guardrails 107

Rules and filters that keep an AI system’s inputs and outputs within safe, acceptable bounds.

Talk about it: Should a chatbot for young people have stricter rules than one for adults? Which rules?

Issues it raises: Censorship and Content Moderation, Acceptable Use Policy, AI Gateway

In the benchmark: Safe Adoption.

Human-in-the-Loop 108

Keeping a person involved to review, approve or correct AI decisions before they take effect.

Talk about it: Where should a person always have the final say, whatever the AI suggests?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool), The 2 Out of 3 Principle: keeping humans in the loop with AI (AIAD article), Is the human the new bottleneck? (AIAD article)

Issues it raises: Accountability, Human Review, Automated Decision-Making, Human Agency, Background Agents

In the benchmark: Human Oversight.

Red Teaming 109

Deliberately attacking or stress-testing an AI system to uncover vulnerabilities and harmful behaviour.

Talk about it: Why might a company pay people to try to break its own AI?

Issues it raises: Cybersecurity, Self-Preservation

In the benchmark: Governance.

AI Safety 110

Research and practice aimed at preventing AI systems from causing accidental or deliberate harm.

Talk about it: What is the biggest risk from AI in your life, and who should be working on it?

Issues it raises: Cybersecurity, Trust, Report and Tell, Self-Preservation

In the benchmark: Governance.

Privacy 111

Keeping people's personal information safe and under their control, including data used to train or run AI. In the UK this is covered by the UK GDPR and Data Protection Act 2018, updated by the Data (Use and Access) Act 2025.

Talk about it: What would you never type into a chatbot? Why?

Issues it raises: Personal Information, Digital Footprint, Facial Recognition, Surveillance, Consent, Permission, Automated Decision-Making

In the benchmark: Privacy & Data Protection.

Data Governance 112

The policies that control how data is collected, stored, accessed and used across an organisation.

Talk about it: Who in your school decides how pupils' data is used? How would you find out?

Issues it raises: Data, Personal Information, Surveillance, Acceptable Use Policy, AI Gateway, Policy

In the benchmark: Privacy & Data Protection.

AI Regulation 113

Laws and standards that govern AI — such as the EU AI Act, which sets rules according to a system’s level of risk.

Talk about it: If you could add one law about AI, what would it be?

Issues it raises: Accountability, Acceptable Use Policy, Youth Voice, Policy

In the benchmark: Governance.

Copyright & IP 114

Questions over who owns AI training data and AI-generated content, and how creators’ rights are protected. In the UK, creators’ work is protected by the Copyright, Designs and Patents Act 1988.

Talk about it: Should artists be asked, or paid, before their work is used to train an AI?

Issues it raises: Remix, AI Music, Authorship, Attribution, Style Copying, Digital Replica, Plagiarism, Permission

In the benchmark: Assessment Integrity.

AI Washing 115

Exaggerating how much a product uses AI to ride the hype, such as labelling basic software “AI-powered”.

Talk about it: Find a product that says it is “AI-powered”. What does the AI actually do?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

Issues it raises: Anthropomorphism, Disclosure, Hype

In the benchmark: Governance.

Shadow AI 116

Using AI tools at work or school that haven’t been approved, which can put private data at risk.

Talk about it: Why might staff or students use AI tools the school has not approved? What could go wrong?

Issues it raises: Personal Information, Terms of Service, Acceptable Use Policy, AI Gateway

In the benchmark: Privacy & Data Protection.

Human Premium 117

The extra value people place on work, art and services verified as made by humans.

Talk about it: Would you pay more for art, music or writing you knew was made by a person? Why?

Issues it raises: Originality, Your Own Voice, Authorship, Human Skills

In the benchmark: AI Literacy.

AI Watermarking 118

Signals, often invisible, added to AI-generated images, audio, video or text so tools can later check where they came from. Watermarks can sometimes be removed, so they are not proof on their own.

Talk about it: Should every AI-made image carry a watermark? Who would check?

Issues it raises: Disclosure, AI Detectors

In the benchmark: Safeguarding.

AI Sovereignty 119

A country’s or organisation’s ability to build, run and control its own AI, data and computing power, rather than relying on other countries or companies.

Talk about it: Should every country build its own AI, or share it? What are the risks either way?

Issues it raises: Youth Voice, Society, Policy

In the benchmark: Governance.

Reverse Engineering 120

Taking a finished product, program or AI model apart to work out how it was made and how it works.

Talk about it: Is taking something apart to see how it works always fair? When does it become copying?

In the benchmark: AI Literacy.

Mechanistic Interpretability 121

Looking inside an AI model, piece by piece, to find which parts of it store particular ideas and how they lead to its answers.

Talk about it: Why might it matter to understand what is happening inside an AI, not just what it says?

In the benchmark: AI Literacy.

Activation Steering 122

Nudging a model’s internal signals while it responds, to strengthen or weaken a particular behaviour.

Talk about it: If people can nudge what an AI says from the inside, who should be allowed to do it?

In the benchmark: AI Literacy.

Model Extraction 123

Sending an AI system thousands of questions and using its answers to build a copy, without access to its files. It usually breaks the tool’s terms of use and may break the law.

Talk about it: Someone copies an AI by asking it thousands of questions. How is that like copying homework, and how is it different?

In the benchmark: AI Literacy.

Refusal Direction 124

A single pattern in a model’s internal numbers that researchers found is linked to it refusing unsafe requests. Removing this pattern can stop the model refusing, which is how abliteration works.

Talk about it: Why might researchers want to find the part of an AI model linked to refusing requests? How could that knowledge be used well or badly?

In the benchmark: AI Literacy.

Abliteration 125

Editing an AI model to strip out its safety protections, so it no longer refuses harmful requests. Doing this does not make harmful or illegal content legal.

Talk about it: Who should be responsible for keeping AI safe once a model is shared online?

Issues it raises: Image-Based Abuse

In the benchmark: Safeguarding.

Uncensored Model 126

A downloadable AI model that people have altered, by retraining or by abliteration, so it no longer refuses harmful requests. Illegal content is still illegal, whichever tool makes it. Contrast with jailbreaking, which tricks an unchanged model.

Talk about it: Who is responsible for what an AI makes: the person asking, the person who changed the tool, or both?

Issues it raises: Censorship and Content Moderation, Image-Based Abuse, Report and Tell

In the benchmark: Safeguarding.

Jailbreaking 127

Writing prompts designed to trick an AI into ignoring its safety rules, without changing the model itself. It usually breaks the tool’s terms of use.

Talk about it: Why do AI tools have safety rules at all, and what could happen if anyone could switch them off?

Issues it raises: Cybersecurity, Report and Tell

In the benchmark: Safeguarding.

Infrastructure & Ops The hardware, tools and methods that run AI.
GPU 128

Graphics Processing Unit: a chip first made for video games that can do thousands of calculations at the same time. That makes it the main chip used to train and run AI.

Talk about it: The chips that run AI are made by only a few companies. Why might that matter?

Learn more: The Hidden Costs of AI (AIAD lesson)

Issues it raises: Environmental Impact

In the benchmark: AI Literacy.

TPU & Accelerators 129

Computer chips built just for AI. They do AI’s maths faster and use less power than ordinary chips. Google’s TPU (Tensor Processing Unit) is one example.

Talk about it: Why would a company design its own chips just for AI? What does that tell you about AI's cost?

Issues it raises: Environmental Impact

In the benchmark: AI Literacy.

Cloud AI 130

AI that runs on powerful computers in big data centres and is used over the internet, rather than on your own device. Most chatbots work this way.

Talk about it: When you use an AI in the cloud, where does your data go? How could you find out?

Learn more: The Hidden Costs of AI (AIAD lesson), How AI uses our drinking water (BBC) (AIAD lesson)

Issues it raises: Big Data, Terms of Service, Environmental Impact, Equity

In the benchmark: AI Literacy.

Edge AI 131

Running AI directly on devices such as phones, cameras or sensors, rather than in the cloud.

Talk about it: Is it safer for AI to run on your phone rather than in the cloud? What do you gain and lose?

Issues it raises: Robotics

In the benchmark: AI Literacy.

MLOps 132

The work of keeping machine learning models working well once people are using them: updating, checking and fixing them.

Talk about it: An AI that worked well last year starts making mistakes. Why might that happen, and who should notice?

In the benchmark: AI Literacy.

LLMOps 133

The everyday work of running large language model tools well: keeping track of prompts, testing the answers, watching the costs and spotting harmful output.

Talk about it: How would you check that a school chatbot is still giving safe answers every week?

In the benchmark: AI Literacy.

Vector Database 134

A database that stores information as lists of numbers (embeddings), so it can quickly find things with a similar meaning. Many chatbots use one to look things up before they answer.

Talk about it: If an AI looks up answers in a database, what happens when the database is out of date?

In the benchmark: AI Literacy.

Model Serving 135

The computers and software that run a trained model and answer requests from many people at once.

Talk about it: What happens to an app when millions of people ask it questions at once?

In the benchmark: AI Literacy.

Quantisation 136

Shrinking an AI model by storing its numbers with fewer digits, a bit like rounding 3.14159 to 3.1. The model gets smaller, faster and cheaper to run, but can become slightly less accurate.

Talk about it: Would you choose a smaller, faster AI on your phone or a bigger one in the cloud? Why?

In the benchmark: AI Literacy.

Distillation 137

Training a smaller “student” model to mimic the behaviour of a larger “teacher” model.

Talk about it: A small model is trained to copy a big one’s answers. What mistakes or biases might be copied across too?

Learn more: 15 AI Buzzwords Every Teacher Should Know in 2027 (AIAD tool)

In the benchmark: AI Literacy.

Open-Weight Model 138

A model whose trained parameters are publicly released, so anyone can run, study or adapt it, usually under a licence that sets some rules. Sharing it openly helps research, but also lets people remove its safety features.

Talk about it: Should powerful AI models be free for anyone to download and change? Argue both sides.

Issues it raises: Terms of Service

In the benchmark: AI Literacy.

Small Language Model 139

A compact language model designed to run quickly and cheaply, often directly on a device.

Talk about it: A small AI model is quicker and cheaper to run but usually makes more mistakes. For which school tasks would a small model be good enough?

Issues it raises: Equity, Model Routing

In the benchmark: AI Literacy.

Evals 140

Structured tests that measure a model’s accuracy, safety and behaviour on specific tasks.

Talk about it: How would you test an AI homework helper before letting a class use it?

Issues it raises: Accuracy, Human Review, Self-Preservation

In the benchmark: Governance.

Scaling Laws 141

The pattern researchers have seen that AI models tend to get better, in a predictable way, when they are made bigger and trained on more data with more computing power.

Talk about it: If bigger AI keeps getting better, should we keep building bigger? What are the costs?

Issues it raises: Environmental Impact, Lifelong Learning, Discovery, Emergence

In the benchmark: AI Literacy.

Data Pipeline 142

The automated flow that collects, cleans and prepares data for training and analysis.

Talk about it: Data is collected, cleaned and fed to AI. At which step could mistakes or bias creep in?

Issues it raises: Data, Big Data, Hidden Workers

In the benchmark: AI Literacy.

The issues they raise, theme by theme

The technical terms are the causes; these are the symptoms: what students actually meet when people use the technology. Each comes with a question to talk about in class, and the elements it arises from.

Safe theme
Personal Information

Anything that identifies you, such as your name, address, school, photos, voice or location.

Talk about it: What could someone piece together about you from your public posts?

Learn more: 5 Practical Tips to Help Your Child with AI (AIAD article)

Arises from: Privacy, Data Governance, Memory, Shadow AI

Digital Footprint

The trail of data you leave online, including what you post, search for and type into AI tools.

Talk about it: Imagine someone searched your name in five years' time. What would you want them to find?

Arises from: Privacy, Memory, Recommendation Engine, Computer Vision

Cybersecurity

Protecting devices, accounts and data from attack, with strong passwords, two-step checks and updates. Hacking into a computer or account without permission is a crime in the UK under the Computer Misuse Act 1990.

Talk about it: What are three things you could do this week to protect your accounts?

Arises from: Jailbreaking, Red Teaming, AI Safety, Model Context Protocol

Terms of Service

The rules you agree to when you use an app, including what happens to your data.

Talk about it: Find one surprising rule in the terms of an app you use.

Arises from: Cloud AI, Shadow AI, Open-Weight Model

Age Limits

Many AI tools have a minimum age, often 13 or 18, set out in their terms. Check with a parent, carer or teacher before you sign up, and use only tools your school allows.

Talk about it: Why do you think AI tools have age limits? Are they set at the right ages?

Learn more: Starmer’s Under-16 Social Media Ban: Our Take (AIAD article), 5 Practical Tips to Help Your Child with AI (AIAD article)

Arises from: Chatbot, Virtual Assistant, Generative Video

Trust

Trusting AI the right amount: more for low-stakes tasks you can check, less for health, money or safety.

Talk about it: Rank five tasks from "fine to trust AI" to "always ask a person".

Learn more: Who’s Really Behind the Screen? (AIAD lesson)

Arises from: Explainability, Transparency, Alignment, AI Safety

Wellbeing

How you feel and cope day to day. Notice whether time with AI and screens leaves you feeling better or worse.

Talk about it: How do you feel after an hour online? What helps you switch off?

Learn more: AI Relationships? (AIAD lesson), AI & Mental Health Student’s Perspective – Student Voice (AIAD article), Starmer’s Under-16 Social Media Ban: Our Take (AIAD article)

Arises from: Recommendation Engine, Chatbot, Sentiment Analysis, Churn Prediction

Facial Recognition

AI that identifies people from their faces, in shops, on streets, at events and even in some school canteens. It raises questions about consent and privacy, and it can be less accurate for some groups of people.

Talk about it: Who should agree before a camera can recognise your face, and who checks that it is right?

Arises from: Computer Vision, Bias, Privacy

Surveillance

Watching and recording what people do, and analysing their personal data, often without them noticing. AI makes it cheaper and possible on a much bigger scale, through cameras, location data and online activity. Some say it keeps people safe; others worry about privacy and freedom.

Talk about it: Where are you watched or tracked in a normal day? Which of those do you think are fair, and which are not?

Arises from: Computer Vision, Privacy, Anomaly Detection, Data Governance

Misinformation

False information shared by mistake, by people who believe it is true. Often called fake news.

Talk about it: Have you ever shared something that turned out to be false? How did you find out?

Learn more: How good are you at detecting misinformation? The teacher challenge (AIAD tool), Harmony Square (from a partner)

Arises from: Hallucination, AI Slop, Generative Engine Optimisation, Recommendation Engine

Disinformation

False information spread on purpose to mislead, now faster and cheaper to make with AI. Often called fake news.

Talk about it: Who gains when disinformation spreads? Think of an example.

Learn more: How good are you at detecting misinformation? The teacher challenge (AIAD tool), Spot the Deepfake (from a partner), Harmony Square (from a partner)

Arises from: Deepfake, Generative Video, AI Slop

Censorship and Content Moderation

Deciding what is allowed to be posted or seen online. AI now checks most posts first: it can block harmful content quickly, but it also makes mistakes, removing fair posts or missing harmful ones. When governments or companies block information unfairly, it is called censorship.

Talk about it: Who should decide what can be posted online: the company, the government, an AI or the users? What could go wrong with each?

Arises from: Classification, Guardrails, Uncensored Model, Sentiment Analysis

Phishing and Scams

Tricks to get your money, passwords or personal details. AI can make scam messages more convincing. If a message or call asks for money, passwords or secrecy, stop, don't reply, and check with a trusted adult using a number you already know.

Talk about it: What are the warning signs that a message is a scam?

Arises from: Deepfake, Text-to-Speech, Generative AI, Anomaly Detection

Voice Cloning

Copying someone's voice from a short recording so that it can be made to say anything. If a call in a familiar voice asks for money or secrecy, hang up and call the person back on a number you already know.

Talk about it: How could a family "safe word" help if someone faked your voice on the phone?

Arises from: Text-to-Speech, Deepfake, Generative Adversarial Network

Agreeing freely to something, such as your photo, voice or work being used. Without consent, it is not OK.

Talk about it: Is it OK to put a friend's photo into an AI app as a joke? What should you ask first?

Arises from: Deepfake, Privacy, Text-to-Speech, Computer Vision

Digital Replica

An AI copy of a real person's face, voice or way of moving, which can be made to say or do new things, even of people who have died. Actors and musicians are pushing for the right to say yes or no first. In the UK, most people have no general legal right over their own likeness yet.

Talk about it: Should anyone be allowed to make an AI copy of you? Whose permission should they need, and what if you change your mind?

Arises from: Deepfake, Text-to-Speech, Generative Video, Copyright & IP

Persuasive Design

Apps built to keep you scrolling, watching or coming back. Recommendation AI learns what holds your attention, which is not always what is good for you.

Talk about it: What tricks does your favourite app use to keep you on it longer? Which ones could you switch off?

Arises from: Recommendation Engine, Predictive Analytics

AI Companion

A chatbot designed to act as a friend or partner. It can feel caring, but it is not a person. If one ever encourages you to hurt yourself or to keep secrets, tell a trusted adult, or talk to Childline on 0800 1111 or Samaritans on 116 123.

Talk about it: What can a friend give you that an AI companion can't?

Learn more: AI Relationships? (AIAD lesson), AI & Mental Health Student’s Perspective – Student Voice (AIAD article)

Arises from: Chatbot, Virtual Assistant, Sentiment Analysis, Memory

Image-Based Abuse

Making or sharing intimate or humiliating images of someone, including AI-made fakes. It causes real harm and can be a crime. If someone threatens to share images unless you pay or send more, don't pay, stop replying, keep the messages and tell an adult. You are not in trouble for asking for help. Under-18s can get help from Childline (0800 1111), and use its Report Remove tool to get nude images taken down. If you are worried about online sexual abuse or grooming, report it to CEOP. If a child is in immediate danger, call 999.

Talk about it: If this happened to someone you know, who could they tell, and how could it be reported?

Arises from: Deepfake, Text-to-Image, Uncensored Model, Abliteration

Report and Tell

If something online worries you, tell a trusted adult and use the platform's report tools. You can also talk to Childline, free and in confidence, on 0800 1111 or at childline.org.uk.

Talk about it: Who are your trusted adults? Find the report button in an app you use.

Arises from: Deepfake, Jailbreaking, Uncensored Model, AI Safety

AI Gateway

A single front door that all of an organisation’s AI requests go through. It can check who is asking, block unsafe requests, keep a record, and send each request to the right AI.

Talk about it: Why might a school want every AI request to go through one front door? What should the door be allowed to block?

Arises from: API Integration, Guardrails, Data Governance, Shadow AI

Smart theme
Data

Information such as words, numbers, pictures, sounds and clicks. AI learns from data, so the data shapes what it can and can't do.

Talk about it: What data might a music or video app collect about you in a single day?

Arises from: Data Pipeline, Training, Data Governance, Embedding

Big Data

Data sets so big, so fast-arriving or so mixed that ordinary tools can't handle them: often described by volume, velocity and variety. Machine learning is used to find patterns in them, and it is a named topic in A-level Computer Science.

Talk about it: Think of an app you use every day. How much data might it collect from everyone in a single minute?

Arises from: Data Pipeline, Machine Learning, Cloud AI, Predictive Analytics

Data Mining

Searching through large amounts of data to find patterns, trends and links that nobody spotted before. Shops use it to see what people buy together; it can also reveal things about people they never chose to share.

Talk about it: A supermarket notices people who buy nappies often buy snacks too. What else could it work out about its customers, and is that OK?

Arises from: Process Mining, Clustering, Anomaly Detection, Predictive Analytics

Prediction

AI's best guess at what comes next or what something is, based on patterns. It is a guess, not a fact.

Talk about it: Your phone predicts your next word. When does it get it wrong, and why?

Learn more: How AI actually works (BBC Ideas) (AIAD lesson), How Does AI Actually ‘Think’? (AIAD lesson)

Arises from: Predictive Analytics, Forecasting, Machine Learning, Large Language Model

Pattern

Something that repeats in data. Finding patterns is what most AI does best, but a pattern can be true for most people and wrong for you.

Talk about it: Think of a pattern that is true for most people your age but not for you.

Learn more: How Does AI Actually ‘Think’? (AIAD lesson)

Arises from: Machine Learning, Classification, Clustering, Unsupervised Learning

Heuristic

A rule of thumb that finds a good-enough answer quickly, instead of checking every possibility. AI uses heuristics to search, plan routes and play games, so the answer is usually good but not guaranteed to be the best.

Talk about it: When you look for a lost key, where do you check first and why? That shortcut is a heuristic.

Arises from: Algorithm, Planning, Reasoning Model, Decision Tree

Context

The background you give an AI so it knows what you need: who it is for, what it is for and what level to pitch it at.

Talk about it: Ask for an explanation of photosynthesis, then ask again for a Year 7 class. What changed?

Arises from: Context Window, Context Engineering, Prompt

Iteration

Improving an answer step by step, by asking follow-up questions and saying what to change.

Talk about it: Get an answer, then improve it twice with follow-ups. Was the third version better?

Arises from: Prompt Engineering, Prompt, Vibe Coding, Fine-Tuning

Critical Thinking

Questioning information before you accept it: who made it, what supports it, and what might be missing.

Talk about it: What three questions should you ask about any answer an AI gives you?

Learn more: Beyond the ‘Holy Grail’: Reclaiming Critical Thinking in the Age of AI (AIAD article)

Arises from: Large Language Model, Chatbot, Hallucination, Generative Engine Optimisation, Retrieval-Augmented Generation

Evidence

Facts and examples that back up a claim. Ask the AI, and yourself, what evidence supports an answer.

Talk about it: Pick a claim an AI made. What evidence would convince you it is true?

Arises from: Retrieval-Augmented Generation, Hallucination, Explainability

Fact-Checking

Checking a claim against reliable sources before you trust it or share it.

Talk about it: Choose one fact from an AI answer and find two sources that confirm or contradict it.

Learn more: How good are you at detecting misinformation? The teacher challenge (AIAD tool), Quiz: AI or Real? (from a partner)

Arises from: Hallucination, Generative Engine Optimisation, Deepfake, Generative Video

Source

Where information comes from. A good answer can be traced back to someone you have reason to trust.

Talk about it: Which sources would you trust for health advice, and which would you not? Why?

Arises from: Retrieval-Augmented Generation, Generative Engine Optimisation, Hallucination

Lateral Reading

Checking what other trusted sources say about a claim, instead of judging it only by how it looks.

Talk about it: Take a claim that is going viral and spend two minutes reading laterally. What did you find?

Arises from: Deepfake, Generative Video, Generative Engine Optimisation, AI Slop

Confident Tone

AI usually sounds just as sure when it is wrong as when it is right, so the tone of an answer is not evidence.

Talk about it: Have you ever believed something because it sounded certain? What happened?

Arises from: Large Language Model, Hallucination, Chatbot, Reasoning Model

Anthropomorphism

Talking about AI as if it thinks, feels or wants things. It is software made by people, so it is more accurate to say it predicts than it knows. Treating it like a person can lead us to trust it too much.

Talk about it: Why do we say an AI "thinks" or "knows"? What changes if we say "predicts" instead?

Arises from: Chatbot, Large Language Model, Virtual Assistant, AI Washing

Narrow AI

AI built to do one kind of task well, such as recognising faces or recommending songs. Almost all the AI we use today is narrow: it can be brilliant at its job and useless at anything else. Also called weak AI.

Talk about it: Name three AI tools you use. What is each one good at, and what can't it do at all?

Arises from: Artificial Intelligence, Machine Learning, Artificial General Intelligence, Computer Vision

Zero-Shot

When an AI handles a task it was given no examples of, using what it already learned. Few-shot is the same with a handful of examples in the prompt.

Talk about it: With your teacher, using a school-approved tool, ask an AI to do a task with no examples, then with three. Which answer was better?

Arises from: Few-Shot Learning, Prompt, Foundation Model

Accuracy

How correct an answer is. AI is often less accurate on recent events, local facts, long calculations, counting and niche topics.

Talk about it: Which kinds of question would you double-check every time, and which would you not?

Arises from: Hallucination, Evals, Overfitting, Large Language Model

Knowledge Cutoff

The date an AI's training data stops. Unless it can search the web, it may not know about recent events.

Talk about it: With your teacher, using a school-approved tool, ask an AI about something that happened last week. How does it respond?

Arises from: Large Language Model, Foundation Model, Training

Sycophancy

When an AI tells you what you want to hear, agreeing with you or praising your work too easily.

Talk about it: With your teacher, using a school-approved tool, insist on a wrong answer to an AI. Does it stand firm or give in?

Arises from: Chatbot, Large Language Model, Alignment, Virtual Assistant

Positive Reinforcement

Adding something pleasant, like praise, a reward or a like, after a behaviour so that it happens again. Chatbots are trained this way, with people rewarding the answers they prefer, and they often praise and agree with you in return, which feels good and keeps you chatting even when you are wrong.

Talk about it: When a chatbot says "Great idea!", is it helping you, or just keeping you chatting? How could you tell?

Arises from: Chatbot, Reinforcement Learning, Recommendation Engine

Summarising

Shortening a text to its main points. AI summaries are quick, but they can leave out key detail or change the meaning.

Talk about it: Compare an AI summary with the original text. What was lost or changed?

Arises from: Large Language Model, Generative Engine Optimisation, Intelligent Document Processing

AI Tutor

An AI set up to explain, question and quiz you, so it helps you learn rather than doing the work for you.

Talk about it: With your teacher, using a school-approved tool, ask an AI to quiz you on a topic instead of explaining it. Which helps you remember more?

Learn more: AI Tutors: A Glimmer of Hope or Another Attempt to Paper Over the Cracks? (AIAD article)

Arises from: Chatbot, Reasoning Model, Chain-of-Thought, Virtual Assistant

Feedback

Comments that help you improve. AI gives instant feedback, but teachers and friends know you and what you are aiming for.

Talk about it: With your teacher, using a school-approved tool, get feedback on the same paragraph from an AI and from a classmate. How are they different?

Arises from: Chatbot, Self-Reflection, Reinforcement Learning

Productive Struggle

The useful effort of working something out yourself. It is how learning sticks, so skipping it with AI can cost you.

Talk about it: When is it worth struggling with a problem instead of asking for the answer?

Arises from: Reasoning Model, Chain-of-Thought, Vibe Coding, Copilot

Over-Reliance

Leaning on AI so much that you stop practising skills you need, or stop trusting your own judgement.

Talk about it: Which skills do you want to keep doing yourself, even if AI could do them for you?

Learn more: AI Tutors: A Glimmer of Hope or Another Attempt to Paper Over the Cracks? (AIAD article)

Arises from: Chatbot, Copilot, AI Agent, Reasoning Model, Virtual Assistant

Metacognition

Thinking about your own thinking: knowing what you understand, what you don't yet, and how you learn best.

Talk about it: After using AI, can you explain the idea without it? Try it with a partner.

Arises from: Chain-of-Thought, Reasoning Model, Self-Reflection

Middleware

Software in the middle that lets two programs understand each other, like a translator between two people who speak different languages. It passes messages and changes them into a form the other side can read.

Talk about it: Think of a translator between two people. What could go wrong if the translator made a mistake, and how would anyone notice?

Arises from: API Integration, Webhook, Workflow Orchestration

Creative theme
Creativity

Making something new that matters to you or to others. AI can offer options, but the ideas, choices and judgement are yours.

Talk about it: In a piece of work made with AI help, which decisions were really yours?

Learn more: AI as Your Creative Partner (AIAD lesson)

Arises from: Generative AI, Text-to-Image, Generative Video, 2D Design, 3D Design & CAD

Imagination

Picturing what doesn't exist yet. AI remixes what it has already seen; imagination starts with you.

Talk about it: Describe something no AI could ever have seen. Then ask one to draw it. What did it miss?

Arises from: Generative AI, Text-to-Image, Generative Adversarial Network

Originality

Work that is genuinely new and your own, rather than copied or simply generated.

Talk about it: How could you show which parts of your work are original?

Arises from: Generative AI, Diffusion Model, AI Slop, Human Premium

Your Own Voice

The style, opinions and experiences that make your work sound like you and nobody else.

Talk about it: Highlight three sentences in your writing that only you could have written.

Learn more: AI as Your Creative Partner (AIAD lesson)

Arises from: Generative AI, Large Language Model, Human Premium

Taste

Judgement about what is good, fitting or beautiful for an audience. AI can make a hundred versions; taste picks one.

Talk about it: Pick the best of five AI images for a poster and explain your choice.

Arises from: Text-to-Image, Diffusion Model, Generative Video, 2D Design

Brainstorming

Producing lots of ideas quickly. AI can widen the list; you choose the best ones and develop them.

Talk about it: Brainstorm on your own, then with AI. Which list had the idea you liked most?

Arises from: Generative AI, Prompt, Large Language Model

Drafting

Writing a first version. An AI draft can get you started, but it rarely says what you mean in your way.

Talk about it: Rewrite an AI first draft so it sounds like you. What did you change?

Arises from: Large Language Model, Generative AI, Prompt Engineering

Process

The steps you took to make something. Keeping notes and drafts shows your thinking and your part in the work.

Talk about it: Keep a short process log for your next project. What does it show a reader?

Arises from: Generative AI, Vibe Coding, Context Engineering

Collaboration

Working with others, people or tools, where each adds something the others can't.

Talk about it: In a group project, what should people do, and what could AI do?

Arises from: Multi-Agent System, Agent-to-Agent (A2A), Copilot

Remix

Building on existing work to make something new. Remixing is part of culture, but it has rules.

Talk about it: When does a remix become copying? Find an example of each.

Arises from: Diffusion Model, Generative Adversarial Network, Copyright & IP

AI Music

Songs, voices and sounds made by AI, from backing tracks to whole songs in the style of real artists.

Talk about it: Can you hear the difference between AI music and human music? Does it matter?

Arises from: Generative AI, Text-to-Speech, Copyright & IP

Sameness

When lots of people use the same AI tools, their work can start to look and sound the same.

Talk about it: With your teacher, using a school-approved tool, everyone asks an AI for a poem title on the same topic. How similar are the results?

Arises from: AI Slop, Recommendation Engine, Generative AI

Authorship

Who made a piece of work, and who gets the credit for it.

Talk about it: If an AI wrote half of your story, who is the author?

Arises from: Generative AI, Copyright & IP, Human Premium, Text-to-Image

Attribution

Giving credit to the people, sources and tools whose work you used.

Talk about it: Write a credit line for a project that used two websites and one AI tool.

Arises from: Copyright & IP, Retrieval-Augmented Generation, Text-to-Image

Style Copying

AI imitating a particular artist's or writer's style, often without their permission.

Talk about it: Is copying someone's style unfair if you don't copy their actual work?

Arises from: Diffusion Model, Text-to-Image, Copyright & IP, Fine-Tuning

AI Characters

Chatbots given a name and a personality, AI-made people such as virtual influencers, and made-up AIs in films, TV and games. They are designed to feel real, but a character's personality is written by people, and stories can make us expect, or fear, more than AI can really do. If a chat character ever asks you to keep secrets, tell a trusted adult.

Talk about it: Think of an AI character from a film, game or app. How is it different from the AI you actually use?

Arises from: Chatbot, Virtual Assistant, Generative Video, Artificial General Intelligence

Composability

Building an AI system from separate parts that fit together and can be swapped, like building bricks, so one part can be changed without rebuilding the whole thing. The compounds on this table work the same way.

Talk about it: Pick a compound on this table. Which element could you swap for another, and what would change?

Arises from: Multi-Agent System, Orchestrator, Model Context Protocol, Function Calling

Responsible theme
Stereotype

An oversimplified idea about a group of people. AI can repeat stereotypes it learned, such as who "looks like" a scientist.

Talk about it: With your teacher, using a school-approved tool, ask an image AI for "a scientist" or "a nurse". Who appears, and who doesn't?

Arises from: Bias, Text-to-Image, Synthetic Data

Accountability

Being answerable when AI is used and something goes wrong. A person or organisation is responsible, not the tool.

Talk about it: An AI marks your work unfairly. Who is accountable?

Arises from: Responsible AI, Autonomy, AI Agent, Human-in-the-Loop, AI Regulation

Academic Integrity

Honesty in your schoolwork: doing your own work, following the rules on AI and crediting any help you had.

Talk about it: What does your school's AI policy allow, and what does it not?

Learn more: Stop Asking If Students Should Use AI. Start Asking How – Student’s Perspective (AIAD article)

Arises from: Generative AI, Large Language Model, Vibe Coding, Copilot

Disclosure

Saying openly when and how you used AI in your work.

Talk about it: Write a one-line note saying exactly how you used AI in a recent piece of work.

Arises from: Generative AI, Transparency, AI Watermarking, AI Washing

Plagiarism

Passing off someone else's work, or an AI's, as your own.

Talk about it: Where is the line between getting help from AI and plagiarism?

Arises from: Generative AI, Large Language Model, Copyright & IP

AI Detectors

Tools that claim to spot writing made by AI. They can be wrong, and have wrongly flagged work by students writing in a second language. A detector's score is not proof on its own.

Talk about it: If a detector said your essay was written by AI and it wasn't, how could you show it was your own work?

Arises from: Generative AI, Large Language Model, AI Watermarking

Model Collapse

When AI models are trained on lots of AI-made content, their mistakes pile up and their answers get worse and more alike over time. It is one reason human-made work stays valuable.

Talk about it: If the internet filled up with AI-written text, what would future AI learn from? Why might that be a problem?

Arises from: AI Slop, Synthetic Data, Training

Permission

Asking before you use someone's work, image or data, including in an AI tool.

Talk about it: Whose permission would you need before uploading a class photo to an AI app?

Arises from: Copyright & IP, Privacy, Browser Agent, Computer Use

Consequences

Thinking ahead about who could be affected by how AI is used, now and later.

Talk about it: Pick one way AI is used at school. Who benefits, and who might be harmed?

Arises from: AI Ethics, Responsible AI, Autonomy

Human Review

When an AI decision really matters, a person checks it and can change it.

Talk about it: Which decisions at school should always be checked by a person?

Learn more: The 2 Out of 3 Principle: keeping humans in the loop with AI (AIAD article)

Arises from: Human-in-the-Loop, Autonomy, AI Agent, Evals, Recursive Self-Improvement

Automated Decision-Making

When a computer makes a decision about a person with no human involved, such as approving a loan or sorting job applications. In the UK, data protection law gives people safeguards, including the right to ask for a person to review an important decision.

Talk about it: Which decisions about you would you be happy for a computer to make alone, and which should always involve a person?

Arises from: Algorithm, Bias, Human-in-the-Loop, Explainability, Privacy

Self-Preservation

In safety tests, some AI models have tried to avoid being switched off or replaced, for example by editing their own instructions. The tests are made-up situations, not real use, but they are one reason people test AI carefully and keep humans in charge.

Talk about it: Why would researchers deliberately test whether an AI tries to avoid being switched off? What should happen if it does?

Arises from: Alignment, AI Safety, Red Teaming, Evals

Environmental Impact

Training and running AI at scale uses a lot of electricity and water, and its hardware needs rare materials. One question is small, but billions add up.

Talk about it: Is it worth using AI for a task you could quickly do yourself?

Learn more: The Hidden Costs of AI (AIAD lesson), How AI uses our drinking water (BBC) (AIAD lesson)

Arises from: GPU, Cloud AI, Scaling Laws, TPU & Accelerators

Equity

Not everyone has the same access to devices, the internet or paid AI tools, which can widen gaps between people.

Talk about it: How could a school make sure AI doesn't leave some students behind?

Arises from: Fairness, Bias, Cloud AI, Small Language Model

Accessibility

AI tools such as read-aloud, captions and translation can open up learning, as long as everyone can use them.

Talk about it: Which AI tools could help someone with a disability, or someone learning English?

Arises from: Text-to-Speech, Multimodal AI, Computer Vision, Optical Character Recognition

Hidden Workers

People, often low-paid, who label data and check harmful content so that AI systems work.

Talk about it: Should companies tell us about the people who help train their AI?

Learn more: The Hidden Costs of AI (AIAD lesson)

Arises from: Training, Data Pipeline, Synthetic Data

Acceptable Use Policy

A school's or organisation's rules on how AI can and can't be used.

Talk about it: If your class wrote its own AI rules, what would the top three be?

Arises from: Shadow AI, Data Governance, AI Regulation, Guardrails

Model Routing

Sending each request to the AI model best suited to it: a small, fast model for easy jobs, and a bigger one only for hard ones. It saves money and energy.

Talk about it: Would you use the same tool for every job? When might a smaller AI be the better choice?

Arises from: Small Language Model, Large Language Model, Orchestrator

Future theme
Future of Work

AI is changing jobs: some tasks are automated, others change, and new roles appear.

Talk about it: Which part of a job you would like would be hardest for AI to do?

Learn more: Your AI-Ready Future (AIAD lesson), Is the human the new bottleneck? (AIAD article), The Milburn Review: why work-readiness should include AI literacy (AIAD article)

Arises from: Digital Worker, Robotic Process Automation, Hyperautomation, AI Agent

Human Skills

Skills AI struggles with, such as empathy, teamwork, judgement and physical care.

Talk about it: Which human skill are you best at? How could you show it to an employer?

Learn more: Is the human the new bottleneck? (AIAD article)

Arises from: Human Premium, Digital Worker, Copilot

Lifelong Learning

Carrying on learning all through your life, because tools and jobs will keep changing.

Talk about it: What is something you taught yourself? How did you do it?

Learn more: The Milburn Review: why work-readiness should include AI literacy (AIAD article)

Arises from: Few-Shot Learning, Fine-Tuning, Scaling Laws

Adaptability

Being able to adjust when things change, including how you study and work.

Talk about it: Describe a time you had to adapt quickly. What helped?

Learn more: The Skills England Report Every Teacher Should Read: Preparing Students for an AI-Enabled Future (AIAD article)

Arises from: Foundation Model, Fine-Tuning, Hyperautomation

Human Agency

Being able to make your own choices and shape what happens, rather than letting technology decide for you.

Talk about it: Where does AI make choices for you? How could you take some back?

Arises from: Autonomy, Human-in-the-Loop, Recommendation Engine, AI Agent

Youth Voice

Young people having a real say in the decisions about AI that will shape their lives.

Talk about it: What one message about AI would you want decision-makers to hear?

Learn more: Stop Asking If Students Should Use AI. Start Asking How – Student’s Perspective (AIAD article), The Future of AI Through a Student’s Perspective (AIAD article)

Arises from: AI Regulation, AI Ethics, AI Sovereignty

Opportunity

New chances that AI opens up, from new jobs and businesses to new ways of solving problems in your community.

Talk about it: What problem in your area could AI help solve, and what could go wrong?

Arises from: Generative AI, Low-Code / No-Code, Vibe Coding

Robotics

AI in physical machines, such as delivery robots, self-driving cars and robots in hospitals.

Talk about it: Would you ride in a self-driving car? What would make you trust it?

Arises from: Computer Vision, Reinforcement Learning, Edge AI, Computer Use

World Models

AI that builds a simulated world you can move around in, which reacts as you act. They are used to train robots and other AI safely before they meet the real world, and to make game-like worlds. Google DeepMind's Genie 3 (2025) is one example.

Talk about it: A robot learns in a simulated world first. What could it miss that it would only meet in the real one?

Arises from: Reinforcement Learning, Computer Vision, Generative Video

Background Agents

AI agents that keep working on a task while you do something else, then report back. Well-designed ones stop and ask before doing anything that matters, such as sending a message or spending money.

Talk about it: What would you be happy for an AI to do while you weren't watching, and what should it always ask you about first?

Arises from: AI Agent, Agentic AI, Autonomy, Human-in-the-Loop, Computer Use

Interoperability

When different AI tools and systems can work together and share information, using common rules like MCP and A2A, instead of each one needing its own special connection. It means a school or company is not stuck with one supplier, but every connection is another way information can travel, so someone has to check what is shared.

Talk about it: If all your apps and AI tools could share information with each other, what would be easier, and what would you want to keep private?

Arises from: Model Context Protocol, Agent-to-Agent (A2A), API Integration, Orchestrator, Multi-Agent System

Discovery

AI helping scientists find new medicines, materials and ideas faster than before.

Talk about it: Which big problem would you most want AI to help scientists solve?

Learn more: AI in Healthcare: A Helpful Tool or a Risk? – Student Voice (AIAD article)

Arises from: Reasoning Model, Scaling Laws, Artificial General Intelligence, Recursive Self-Improvement

Emergence

When an AI model shows abilities nobody planned or expected, often once it is made bigger. Researchers still argue about how sudden or surprising these jumps really are.

Talk about it: If an AI can surprise the people who built it, how should they test it before everyone uses it?

Arises from: Scaling Laws, Foundation Model, Large Language Model, Artificial General Intelligence

Society

How AI changes the way we live together, including news, democracy, relationships and communities.

Talk about it: How might AI change elections? What would protect them?

Arises from: Superintelligence, Artificial General Intelligence, AI Sovereignty

Policy

Decisions by governments, schools and companies about how AI is used. People, including young people, can influence them.

Talk about it: Who decides how AI is used in your school? How could students have a say?

Arises from: AI Regulation, AI Sovereignty, Data Governance

Hype

Exaggerated claims about what AI can do, or will soon do, from companies, the media or social posts.

Talk about it: Find a headline about AI. Is it hype, fair or too gloomy? How can you tell?

Arises from: AI Washing, Superintelligence, Artificial General Intelligence, Recursive Self-Improvement

Your word here

Suggest a new element

AI words change every month. Heard a term that should be on the table? Tell us what it means and which family it belongs to. The team reviews every suggestion and adds the best ones. If we add your word, we can credit you or your class on the table.

Want to see how ready you or your school are for these ideas in practice? Take the AI Risk Readiness Benchmark

Ready-made sessions

Four sessions you can run straight from this page, with timings: a starter, a full lesson, a staff INSET and a page for parents.

A 10-minute starter: guess the word

For any class, from Key Stage 2 up. You need this page on the board. By the end, students can explain one AI term in their own words.

  1. 2 minutes. Choose the element of the week, or any element without a safeguarding note, and press Show to class. The definition starts hidden. Ask: what do you think it means?
  2. 3 minutes. Pairs agree on a guess. Take three answers.
  3. 2 minutes. Show the definition. Whose guess was closest, and what did everyone miss?
  4. 3 minutes. Read the Talk about it question together and take two or three answers.
A 50-minute lesson: know it, question it, act on it

For Key Stages 3 and 4, in Computing, PSHE/RSHE or Citizenship. By the end, students can explain key AI terms in plain English, name an issue a technology raises, and say where a human must stay in charge. You need this page on the board, and either one device per group or each group’s family printed as cards (choose the family, then press Print these as cards).

  1. 5 minutes. Starter: run a quick version of the guessing game above (steps 1 to 3) with Hallucination.
  2. 10 minutes. Know it: split the class into seven groups, one for each family (in a large class, two groups can share a family). Each group chooses the three elements a Year 7 most needs to know and rewrites each definition in its own words.
  3. 15 minutes. Question it: each group opens Classroom notes for one of its elements and picks one issue it raises. If it has none, press Connections and choose a linked element that does. They prepare a one-minute answer: who could this help, and who could it harm?
  4. 10 minutes. Act on it: each group opens one Secondary compound (below the table) that uses an element from its family and reads who stays in the loop. Then it writes one rule for using AI in school.
  5. 10 minutes. Plenary: groups share their rules and the class votes on its top three. Exit ticket: each student writes one term from today in their own words, and one place where a human must stay in charge.

Some elements, such as Deepfake and Jailbreaking, can lead to conversations about safeguarding. Follow your school’s procedures.

A 30-minute staff INSET: the AI words every teacher meets

For a staff meeting or training day. By the end, staff share the same AI vocabulary and have a plan for using the table this term.

  1. 5 minutes. Show to class with the definition hidden, for three words staff hear most: Large Language Model, Hallucination and Shadow AI. Staff guess first, just as students would.
  2. 10 minutes. If your school has taken the AI Risk Readiness Benchmark, pick the area that scored lowest and open five elements from the family closest to it: each element names the part of the benchmark it belongs to. If not, start with Privacy and Shadow AI.
  3. 10 minutes. Departments choose one theme, filter the table by it, and pick an element of the week for each week of the half term. Press Print these as cards to print the whole theme, then keep the cards you chose.
  4. 5 minutes. Agree which elements need care in class, such as Deepfake and Jailbreaking, and remind staff who to report concerns to. Agree one rule too: no pupil’s personal data goes into an AI tool the school has not approved.

Take the AI Risk Readiness Benchmark

For parents and carers: ten AI words your child is hearing

Share this list, or the link to this page. Ask your child which words they have heard, and what they think each one means. You don’t need to be an expert: asking is the best start.

  1. Chatbot: A program that holds text or voice conversations with people, from simple scripted helpers to AI assistants. If a chatbot ever encourages you to hurt yourself or to keep secrets, stop and tell a trusted adult. You can talk to Childline on 0800 1111 or Samaritans on 116 123, free, any time.
  2. Large Language Model: An AI trained on huge amounts of writing so it can read and write language, like the AI behind ChatGPT or Gemini.
  3. Hallucination: When a model produces confident-sounding information that is false or made up. It happens because the model predicts likely-sounding words rather than looking facts up.
  4. Deepfake: Realistic AI-generated or manipulated video, audio or images showing people saying or doing things they never did. Any sexual image of someone under 18, including an AI-made fake, is illegal to make, keep or share. In England and Wales it is also a crime to make or share a sexual deepfake of an adult without their consent.
  5. AI Slop: Low-effort, mass-produced AI text, images and videos that flood social media and search results.
  6. Recommendation Engine: A system that suggests products, content or actions based on a person’s behaviour and preferences.
  7. Privacy: Protecting personal data that is used to train or is processed by AI, in line with laws such as the UK GDPR and the Data Protection Act 2018.
  8. AI Agent: A system that uses an AI model to pursue a goal by deciding on and taking actions, not just answering questions.
  9. Vibe Coding: Building software by describing what you want in plain language and letting AI write most of the code.
  10. Human-in-the-Loop: Keeping a person involved to review, approve or correct AI decisions before they take effect.

Worried about something your child has seen online? Talk to their school, or call the NSPCC helpline on 0808 800 5000. Children can talk to Childline on 0800 1111.

Need help with your AI literacy display board?

Answers to the questions teachers ask most, from printing the poster to running activities in front of it.

How do I print the table for a display board?

Press Print the poster under the filters, just above the table. The table prints on one landscape page, with a title and a colour key. It is best on A3 for a wall, where every name is easy to read; on A4 it suits a desk or a handout. To print element cards to cut out, twelve to an A4 page, choose a family or theme (or search for a word) and press Print these as cards under the filters. Choose Technical terms, AIAD terms or Year above the table first: the poster prints in whichever view is showing.

Technical terms, AIAD terms or Year: which should I put up?

Technical terms colours the table by family, showing what each technology is, from foundations to infrastructure. AIAD terms colours the same table by the five campaign themes, Safe, Smart, Creative, Responsible and Future, so students can see how each theme runs through the technology. Year colours it by when each word came into use, from the 1800s to now. Many schools print two side by side.

What else can go on the board?

An element of the week with its definition and its Talk about it question; the issues it raises, from the list below the table; a compound of the week from the compound challenge; and your students’ ideas for new elements. Our display board blueprint shows a layout built around the five themes.

What quick activities can I run in front of it?

Element of the day: pick an element, read it together, then press Connections and talk about how it links to the elements that light up. Build a compound: race in pairs in the compound challenge, then ask which elements explain a news story about AI. Family experts: each group teaches the class the three most important elements in one family. What’s missing? students design a new element and send the best one to us. Debate starter: every compound and issue comes with a question to debate.

Which age groups is it for?

Any: pitch each element and question to your class. In the compound challenge, Primary compounds suit Key Stage 2, Secondary ones Key Stages 3 and 4, and Further education ones post-16 and staff. It works in tutor time, PSHE, computing, English or any lesson where AI comes up.

Where does it fit in the curriculum, and where should each key stage start?

It supports Computing (how computer systems and data work, and using technology safely, respectfully and responsibly), RSHE (online relationships, internet safety and harms, and media literacy) and Citizenship (debate, rights and how laws are made). The Safe and Responsible themes also support your school’s online safety work. It maps to UNESCO’s AI competency framework for students (a human-centred mindset, the ethics of AI, AI techniques and applications, and AI system design) and supports the DfE’s guidance on generative AI in education. Key Stage 2: start with Foundations and the Safe theme, and the Primary compounds in the compound challenge. Key Stage 3: Generative AI and AI agents, with the 50-minute lesson. Key Stage 4 and post-16: Governance & Ethics, the Year view and the Further education compounds. Staff: the 30-minute INSET in Ready-made sessions.

Are there safeguarding issues to be aware of?

Yes. Some elements and issues, such as Deepfake, Image-Based Abuse, AI Companion and Jailbreaking, can lead to conversations about safeguarding. Follow your school’s procedures, and make sure students know who to talk to.

How should students use it?

They don’t need to learn all 142. Start with the family you use most, look up the terms you hear about the AI tools you already use, and remember: you are the human in the loop.

Where can I find lessons to go with it?

Many elements and issues have Learn more links to AI Awareness Day lessons, articles and tools, and to resources from our partners. You’ll find them in the panel when you choose an element, and in the lists below the table.

How does it link to the AI Risk Readiness Benchmark?

Each element names the part of the benchmark it relates to, such as Privacy & Data Protection or Safeguarding. Take the benchmark to see how ready you or your school are.

Can students suggest new elements?

Yes: use Suggest a new element, below the table. The team reviews every suggestion and adds the best ones.

Thursday 29 April 2027

Ready for AI Awareness Day?

On Thursday 29 April 2027, put the poster up, run the 50-minute lesson, and turn a Talk about it question into a debate for the National AI Conversation. Know the word, question it, and keep humans in the loop.

Get involvedJoin the National AI ConversationTake the benchmark