AI words in plain English.
Short, everyday explanations of the words you will meet when you use AI. No jargon in the answers.
This month: what is an AI agent?
An ordinary AI assistant writes words. An AI agent can also do things: search, open pages, fill in forms or send messages for you. That is handy, but it means a mistake can have real effects, such as an email sent to the wrong person or a booking made that you did not intend.
What to do: start with small, low-risk jobs. Ask to see what it plans to do before it does it. Never give an agent your passwords or payment details, and read what it did afterwards.
AI agent
An AI that can take steps on your behalf, such as searching, filling in a form or sending an email, not only writing text. The more it can do, the more carefully you should check what it did.
AI assistant
A program that reads what you type and writes a reply, such as ChatGPT, Gemini, Copilot or Claude. It predicts likely words. It does not look things up unless it has a search tool, and it does not know when it is wrong.
Bias
A leaning in an AI's answers that comes from the information it learned from. It can make answers unfair to some people. This is one reason AI must not decide things about people.
Chatbot
A program you talk to by typing. Some chatbots are AI assistants. Others only follow fixed scripts.
Context
Everything the AI can see in the conversation so far. It cannot see anything you have not given it, and long conversations can push early details out of view.
Copilot
Microsoft's family of AI assistants. Some versions are built into Microsoft 365 and can see your work files if your organisation allows it.
Data privacy
Keeping personal and sensitive information to yourself. Before you put anything into an AI tool, ask whether you are happy for that tool to see it.
Deepfake
A fake photo, video or voice recording made by AI to look like a real person. Scammers use them. If a call or video asks for money, check by ringing the person on a number you already have.
Generative AI
AI that makes new text, pictures, audio or video, rather than only sorting or predicting.
Hallucination
When an AI states something that is false as if it were true, such as an invented fact, number, book or phone number. It happens even when the AI sounds sure.
Human in the loop
A person who checks the AI's work before it is used. In these courses, that person is you.
Image generator
An AI that makes pictures from a written description. The pictures can look real but show things that never happened.
Input
What you type or paste into an AI tool. Treat it as something you are sharing.
Large language model (LLM)
The kind of AI behind most chat assistants. It learned patterns from a huge amount of text and uses them to write likely replies.
Memory
A feature in some AI tools that remembers things about you between chats. Check the settings, and do not let it store private details.
Model
The AI engine itself. Different models are better at different things, and they are updated often.
Output
What the AI writes back. Always check it before you use it.
Phishing
A fake message made to look like it is from a real bank, shop or agency, to trick you into clicking a link or giving details.
Placeholder
A marker such as [Name] that you leave in a prompt or a draft, then fill in yourself. It lets you get help with wording without sharing private details.
Prompt
The instructions you give an AI assistant. A good prompt says who it is for, gives the facts, says what you want back and what to do when something is missing.
Prompt library
A collection of ready-made prompts you can copy and adapt, such as the JP226 Prompt Library.
Retrieval
When an AI looks up documents or web pages and uses them in its answer. It can still misread them, so check the source.
Scam
A trick to get your money, passwords or personal details.
Search tool
A feature that lets an AI look things up on the web. It makes answers more current, but it can still choose a poor source.
Sensitive information
Details that could cause harm if shared, such as health, finances, passwords, children's details or security codes.
Summarise
To shorten a document into its main points. AI can do this well, but it can leave out or change details, so compare the summary with the original.
System prompt
Hidden instructions from the company that made an AI tool, which shape how it behaves. You usually cannot see them.
Token
A small piece of text, often part of a word, that an AI reads and writes. Tools often have limits measured in tokens.
Training data
The text, pictures and other material an AI learned from. It does not contain your private chats unless the tool's settings say so.
Verification
Checking a claim against a trustworthy source you reach yourself, such as an official website, a bill or a record.