The AI glossary
94 words from AI news, explained simply, each with a one-line version for young readers and what it means for people.
A
- AI agentAn AI agent is an AI that doesn't just tell you how — it goes and does the steps for you.
- AI alignmentAlignment is making sure an AI wants what we actually want, not a weird version of it.
- AI and copyrightAI and copyright is the big argument about whether AI can learn from people's work without asking, and who owns what AI makes.
- AI assistantAn AI assistant is a helper app that can explain things, help you write and answer questions.
- AI detectorAn AI detector tries to guess if a computer wrote something — and it often guesses wrong.
- AI governanceAI governance is all the rules people make about how AI should be used, from countries down to classrooms.
- AI literacyAI literacy is knowing how AI works well enough to use it smartly and safely.
- AI memoryAI memory is when the assistant remembers things you told it last time.
- AI modelA model is the "brain file" an AI app uses, made by training it on lots of examples.
- AI safetyAI safety is everything people do to make sure AI helps and doesn't hurt.
- AI washingAI washing is a company pretending its product uses amazing AI when it doesn't really.
- AI watermarkingA watermark is a hidden "made by AI" stamp inside a picture or sound.
- AlgorithmAn algorithm is a recipe a computer follows, step by step.
- Algorithmic biasAlgorithmic bias is when an AI treats some people unfairly because it learned from unfair or incomplete examples.
- Algorithmic impact assessment (AIA)An impact assessment is thinking hard about who might get hurt before you switch an AI on.
- API (application programming interface)An API is a doorway that lets one app talk to another app's brain.
- Artificial general intelligence (AGI)AGI is the idea of an AI that could learn anything a person can — it doesn't exist yet, and people argue about whether it will.
- Artificial intelligence (AI)AI is software that has learned from lots of examples so it can guess good answers, a bit like a very fast pattern-spotter.
- Artificial Intelligence and Data Act (AIDA)AIDA was a plan for Canada's first AI law; it ran out of time in Parliament and never passed.
- Automated decision-makingAutomated decision-making is when a computer, not a person, decides something about you.
- AutomationAutomation is machines or software doing jobs that people used to do by hand.
B
C
- Chain of thoughtChain of thought is the AI writing out its steps, like you do in math class.
- ChatbotA chatbot is a computer you can text with that texts back.
- ComputeCompute is the amount of computer muscle an AI needs to learn and to answer.
- Computer visionComputer vision is teaching computers to see and name what's in a picture.
- Content provenance (C2PA)Content provenance is a picture's ID card that says where it came from and what was changed.
- Context windowThe context window is the AI's short-term memory: how much of the chat it can keep in mind at one time.
D
- Data centreA data centre is a giant warehouse of computers where the internet and AI actually run.
- Data labellingData labelling is people putting name tags on examples so the AI knows what it's looking at.
- Data retentionData retention is how long a company keeps what you typed or uploaded.
- Deep learningDeep learning is a stack of tiny decision-makers: the first ones spot edges, the next ones spot shapes, and the last ones say "that's a cat".
- DeepfakeA deepfake is a fake video or voice of a real person made by AI — it can look and sound real, so check before you believe or share.
- Diffusion modelA diffusion model starts with TV static and cleans it up, bit by bit, until a picture appears.
- DistillationDistillation is a big AI tutoring a smaller AI until the little one is almost as good.
E
- EmbeddingAn embedding is a secret number-address for an idea, so ideas that mean the same thing live close together.
- EU AI ActThe EU AI Act is Europe's big rulebook for AI: some uses are banned and risky ones have to follow strict rules.
- ExplainabilityExplainability is being able to answer "why did the AI decide that?"
F
- Facial recognitionFacial recognition is a computer recognising who you are from your face.
- Few-shot promptingFew-shot prompting is showing the AI two or three examples so it gets the idea.
- Fine-tuningFine-tuning is giving an AI extra lessons in one subject so it gets really good at it.
- Foundation modelA foundation model is the big base that lots of different AI apps are built on top of, like the foundation of many houses.
- Frontier modelFrontier models are the newest, most powerful AIs — the ones at the edge of what's possible right now.
G
- Generative AIGenerative AI is a computer that makes new stuff, like stories or pictures, after studying millions of examples.
- GPU (graphics processing unit)A GPU is a chip that can do thousands of tiny maths problems at once, which is exactly what AI needs.
- GroundingGrounding is the AI showing where it got its answer, like a bibliography.
- GuardrailsGuardrails are the safety rails that keep an AI from saying or doing harmful things.
H
I
J
K
L
M
- Machine learningMachine learning is teaching a computer with examples instead of instructions, like learning to ride a bike by practising, not by reading a manual.
- Misinformation and disinformationMisinformation is wrong info shared by mistake; disinformation is wrong info shared on purpose to trick you.
- Model Context Protocol (MCP)MCP is like a universal charger cable that lets any AI plug into lots of different apps.
- Multimodal AIMultimodal AI can read, look, listen and talk, not just type.
N
O
P
- ParametersParameters are the billions of little settings inside an AI that get tuned while it learns.
- Personal informationPersonal information is anything that could tell someone who you are or things about you.
- PIPEDAPIPEDA is Canada's rulebook for how businesses must treat your personal information.
- Pre-trainingPre-training is the AI's giant reading marathon before it learns any manners.
- PromptA prompt is the message you send the AI to tell it what you want.
- Prompt engineeringPrompt engineering is getting really good at asking the AI questions so it gives you what you actually need.
- Prompt injectionPrompt injection is sneaking secret orders into a web page so an AI reading it gets tricked.
Q
R
- Reasoning modelA reasoning model is an AI that "shows its work" in its head before answering.
- Red teamingRed teaming is people trying hard to break an AI on purpose, so it can be fixed before everyone uses it.
- Reinforcement learningReinforcement learning is learning like a video-game player: try, score points, try again smarter.
- Reinforcement learning from human feedback (RLHF)RLHF is people giving an AI thumbs up or thumbs down until it learns which answers are helpful.
- Reskilling and upskillingUpskilling is getting better at your job; reskilling is learning a new one.
- Responsible AIResponsible AI means using AI in ways that are fair, safe and honest.
- Retrieval-augmented generation (RAG)RAG is when the AI looks things up in a book first, then answers — like an open-book test.
S
- Small language model (SLM)A small language model is a pocket-sized AI that can live right on your phone.
- Sovereign AISovereign AI is a country making sure it can run its own AI, not only borrow someone else's.
- Speech recognitionSpeech recognition is the AI typing what you say.
- Supervised learningSupervised learning is studying with an answer key.
- Synthetic dataSynthetic data is pretend data made by a computer so an AI can practise without using real people's information.
- System promptThe system prompt is the secret rule sheet the AI reads before it talks to you.
T
- TemperatureTemperature is the AI's "surprise me" dial: low is careful, high is wild.
- Text-to-imageText-to-image is typing a description and getting a brand-new picture.
- Text-to-speech (TTS)Text-to-speech is a computer reading out loud.
- TokenTokens are the puzzle pieces an AI cuts words into before it reads them.
- Tool useTool use is the AI grabbing a calculator or a search engine when it needs one.
- Training dataTraining data is the giant pile of examples an AI studies before it can do anything.
- TransformerA transformer is the clever design that lets an AI look at all the words in a sentence together to figure out what they mean.
U
V
- Vector databaseA vector database is a library sorted by meaning, so the AI can grab the right pages fast.
- Vibe codingVibe coding is making an app by telling the AI what you want instead of typing the code yourself.
- Voice cloningVoice cloning is AI copying someone's voice — so a phone call that sounds like Grandma might not be Grandma.