What AI tools are and how they produce answers
- Video
- 3 min
- Reading
- 15 min
- Exercise
- 10 min
- Total
- 27 min
Educational content, not legal advice.
In this lesson you will
- Recognize an AI tool at work, including tools that score, rank or sort without calling themselves AI.
- Explain in plain words how a chat assistant produces an answer, and why fluent is not the same as true.
- Describe where your words go when you type into an AI tool, and why the account type matters.
- Choose tasks where AI helps and tasks where you rely on your own checking and judgment.
Video lesson
Read the transcript
Welcome to Valdra Academy. This is lesson one of AI Literacy at Work. In the next few minutes you will learn what an AI tool is, how it produces an answer, and why that changes the way you use it.
AI tools are already part of most jobs. Chat assistants draft emails and summarize documents. Meeting tools write notes. Other software ranks, scores or sorts things for you, often without a label that says AI. And some tools make images, voices or video. In this course, an AI tool is any tool that learns patterns from data and uses them to write, predict, rank or decide.
So how does a chat assistant produce an answer? It was trained on a huge amount of text. From that text it learned which words tend to follow which. When you ask a question, it predicts a likely next word, then the next, and the next. The result reads well, because good writing was in its training. But fluent is not the same as true. The tool does not check facts the way a person does. It produces what sounds likely. Most of the time, what sounds likely is also right. Sometimes it is not, and nothing in the answer tells you which.
That explains three things you will notice. Ask the same question twice and you can get two answers. It might not know recent events, because its training stopped at a certain date. And it knows nothing about your company unless someone connects it to your files, or you paste the information in. So when it seems to know your rules, check whether it is guessing.
Which brings us to your words. What you type is sent to the tool’s provider. Two questions decide how risky that is. First, is your text stored? Second, is it used to train the model? On a personal account, the answer can be yes to both. On an account your organization approved, both are usually settled by contract. That is why your organization approves certain tools, on business terms, for work data.
The law cares about this too. Where PIPEDA applies, personal information has to be protected by safeguards that suit how sensitive it is. PIPEDA Sch.1 cl. 4.7 And in the European Union, the AI Act says that providers and deployers of AI take measures to support the AI literacy of their staff. AI Act Art. 4(1), as replaced by Regulation (EU) 2026/1744; Art. 113, third paragraph, point (a) This course is one of those measures. At our fictional property manager, Quillfield, every employee takes it in their first month.
Recap: an AI tool predicts what sounds likely. Treat its answer as a draft, and know where your words go. In the next lesson: where AI goes wrong, and how to check it.
Why this course exists
Most people now use AI at work, often without calling it that. A writing assistant suggests the next sentence. A meeting tool writes the notes. A chat assistant drafts a reply to a customer. These tools save real time. They also make a new kind of mistake, and they send your words somewhere you might not expect.
This course is for everyone who uses AI tools at work, not for specialists. You will not need any technical background. By the end you will know how these tools produce answers, where they go wrong, what never goes into them, how to use them fairly and safely, and when to stop and ask.
What we mean by an AI tool
In this course, an AI tool is any tool that learns patterns from data and uses them to write, predict, rank or decide. That is a plain working test, not a legal definition.
You will meet AI in four main places at work.
| Where | Examples |
|---|---|
| Chat assistants | Drafting emails, summarizing a long document, answering a question, translating |
| Features inside software you already use | Meeting notes, suggested replies, "smart" search, writing help in a word processor |
| Tools that score, rank or sort | A tool that ranks job applications, flags unusual invoices, or sorts maintenance requests by urgency |
| Tools that make images, audio or video | A generated photo for a flyer, a cloned voice, a video avatar |
The last two are easy to miss. A scoring tool might not look like AI at all: it just shows a number or puts items in an order. Yet it learned that order from data, and it can be wrong in the same ways.
The EU AI Act has its own, more detailed definition of an AI system. In short: a machine based system that infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions. AI Act Art. 3(1), (3) and (4) You do not need to apply that definition yourself. Your organization decides which tools are in scope. Your job is to recognize when a tool is doing this kind of work.
How a chat assistant produces an answer
A chat assistant runs on a large language model. Here is what that means in plain words.
- Training. The model was shown a huge amount of text: books, websites, code and more. From it, the model learned which words tend to follow which, in which contexts.
- Prediction. When you type a question, the model predicts a likely next word, then the next one, and so on, until the answer is complete.
- Polish. The model was then tuned to sound helpful and polite, and to follow instructions.
The result is fluent, well organized text. But fluent is not the same as true. The model does not look up a fact and check it the way a careful person would. It produces what sounds likely, based on patterns. Most of the time, what sounds likely is also correct. Sometimes it is not, and the wrong answer sounds just as confident as the right one.
Some tools are connected to a web search or to your company’s documents. They can then quote real sources. That helps, but the tool can still misread a source, mix two sources together, or cite a page that does not say what it claims. Lesson 2 shows you how to check.
Three things you will notice
Once you know the tool predicts likely words, three everyday puzzles make sense.
- The same question gives different answers. There is a little randomness built in, so each answer is a new prediction. If two answers disagree, that is a signal to check, not to pick the one you like.
- It might not know recent events. Training stops at a certain date. Unless the tool searches the web, it might not know about a new rule, a new price or last week’s news. It might not tell you that it does not know.
- It knows nothing about your company unless someone connects it to your files or you paste the information in. When it seems to know your policies, check whether it is guessing.
Where your words go
When you type into an AI tool, your text is sent to the company that runs the tool, usually over the internet, and processed on its computers. What happens next depends on the account and the settings.
| Question | Personal or free account | Account your organization approved |
|---|---|---|
| Is your text stored? | Often yes, sometimes for a long time | Set by the contract, often short or off |
| Is it used to train the model? | Can be yes, unless you switch it off | Usually no, by contract |
| Can your employer see or delete it? | No | Yes, in most business plans |
| Who agreed to the terms? | You, as a private person | Your organization, after a review |
These are typical answers; the terms of each tool and account decide.
This is the single most important practical fact in this lesson. The same tool can be safe for work data on a business account and unsafe on a personal one. That is why your organization approves certain tools for certain uses. Lesson 3 covers what never goes into any tool, approved or not.
What AI is good at, and where it is weak
| Usually good at | Usually weak at |
|---|---|
| A first draft of an email, a notice or a summary | Exact facts, figures, dates and quotes |
| Rewording text to be shorter, plainer or more polite | Knowing your company’s own rules and context |
| Brainstorming options and questions | Recent events and changes in rules |
| Explaining a general idea in simple words | Telling you when it is unsure |
| Translating everyday text | Legal, medical, financial or safety advice you can rely on |
A good rule of thumb: use AI to help you start, and use your own judgment to finish.
Why the law cares
AI tools touch two things the law protects: information about people, and decisions about people. In Canada, where the federal private sector privacy law (PIPEDA) applies, personal information has to be protected by security safeguards that suit how sensitive it is. PIPEDA Sch.1 cl. 4.7 What staff type into an AI tool is part of that picture.
In the European Union, the AI Act says providers and deployers of AI systems take measures to support the AI literacy of the staff and others who operate or use AI on their behalf, and it sets no specific level that each person has to reach. AI Act Art. 4(1), as replaced by Regulation (EU) 2026/1744; Art. 113, third paragraph, point (a) A short course like this one, plus clear rules and a person to ask, is the kind of measure employers use. Lesson 6 explains more.
Three habits to start today
- Treat every answer as a draft. You are the author of anything you send, file or decide.
- Know which tools your organization has approved, and for what kind of information.
- When something feels off, ask. A two minute question to your manager or IT is cheaper than a mistake that reaches a customer.
Quillfield maps the AI its staff already use
Fictional organization. Quillfield Property Management Inc. is invented for this course. Any resemblance to a real company is a coincidence.
Quillfield manages about 2,400 rental apartments in Nova Scotia, New Brunswick and Montreal. It has about 90 staff: leasing agents, property managers, maintenance coordinators, a small accounts team and tenant services. Its head office is in Halifax.
Before rolling out this course, the operations director asked a simple question at a staff meeting: "Which AI tools do you use, even a little?" She promised that nobody would get in trouble for the answer. Here is what came back in one week.
| Tool or feature | Who uses it | What it does | Is it AI? |
|---|---|---|---|
| Free chat assistant, personal accounts | Leasing and tenant services | Drafts emails and listing text | Yes |
| Meeting notes in the video call software | Managers | Writes a summary after each call | Yes, switched on by the vendor |
| Urgency sorting in the maintenance system | Maintenance coordinators | Sorts repair requests | Yes, learned from past requests |
| Applicant score in the tenant screening service | Leasing | Gives each rental applicant a score | Yes, and it shapes decisions about people |
| Spreadsheet formulas in the rent roll | Accounts | Adds up rent and fees | No, fixed rules a person wrote |
| Translation feature in the email program | Tenant services, Montreal | Translates tenant emails | Yes |
What the team learned. Half the AI in use was not "an AI tool" in anyone’s mind. The urgency sorting and the applicant score just looked like software. The meeting notes had been switched on by a vendor update. And the most common tool, the free chat assistant, was being used on personal accounts, so tenant names and unit numbers were going to a provider under terms no one at Quillfield had read.
What changed. Quillfield bought a business licence for one chat assistant, with storage of chats limited and training on its data switched off by contract. It asked staff to move work tasks to that tool. It added the applicant score and the urgency sorting to its list of AI tools, with a named owner for each. And it made this course part of every new employee’s first month.
Why it matters for you. You cannot use a tool carefully if you do not notice it is AI. The first skill in this course is simply to see it.
Spot the AI in your own day
Think about one normal working day in your job, or use this one: you are a tenant services representative. In one morning you (a) read an email that your email program translated from French, (b) use a chat assistant to draft a reply to a noise complaint, (c) look up a tenant’s balance in the accounting system, and (d) see that the maintenance system has marked a leak as "urgent".
For each of the four moments, write a few words:
- Is AI involved? How can you tell?
- If it is, what is the most likely way it could be wrong?
- Is any information about a person going into the tool?
Show the model answer
- Translated email: yes, AI. Machine translation learned from data. It can miss tone or turn a polite request into a rude one, and it can mistranslate a number or a date. The tenant’s email, with their name and maybe their unit, went into the tool. Check that the translation feature is one your organization has approved.
- Drafted reply: yes, AI. The chat assistant predicts likely words. It could invent a rule ("quiet hours start at 9 pm") that is not in the lease, or promise something you cannot deliver. If you pasted the complaint in with names, personal information went into the tool, so it belongs in the approved tool only.
- Tenant balance: probably not AI. An accounting system that adds up charges follows fixed rules. Still, the balance is personal information, so it stays out of any chat assistant.
- Urgent leak: likely AI. The urgency label probably comes from a model that learned from past requests. It can be wrong in both directions: a real emergency labelled routine, or a small drip labelled urgent. A person still decides what happens.
Why this is the strongest answer
The strongest answers do two things. They notice AI even when it does not announce itself, as in the translation and the urgency label. And they separate two different questions: "could the output be wrong?" and "is information about a person going in?". The first question is about checking (lesson 2). The second is about what never goes into a tool, and which tools are approved (lesson 3). Most mistakes at work come from forgetting one of the two.
Before you use an AI tool: a one page checklist
Six questions to ask before you use AI for a work task: is the tool approved, what goes in, who checks the output, who is affected, do you need to say AI was used, and whom to ask.
Check your understanding
Sources cited in this lesson
- AI Act Art. 3(1), (3) and (4) official text current to 2026-07-27, checked on 2026-10-29
- PIPEDA Sch.1 cl. 4.7 official text current to 2026-09-30, checked on 2026-10-24
- AI Act Art. 4(1), as replaced by Regulation (EU) 2026/1744; Art. 113, third paragraph, point (a) official text current to 2026-07-27, checked on 2026-11-01