Skip to content

Free sample · Module 1: What AI is and what it is not · 4 min read

What counts as AI at work

You already use AI at work, whether or not you chose to. The filter that moves junk email out of your inbox is AI. So is the tool that turns a recorded meeting into a transcript, the suggested reply under an email, the translation button in your browser, and the chat assistant that drafts a first version of a report. Some of these arrived quietly inside software your company already had. That is why AI literacy starts with a plain question: which of the tools in front of me are AI, and what kind?

A working definition

An AI system is software that works out its own way from input to output. Nobody wrote the rule it follows for your particular case. It was built by showing it very large numbers of examples, and from those it learned patterns that it now applies to new input. European law uses a similar idea. The EU AI Act describes an AI system as a machine-based system that infers, from the input it receives, how to generate outputs such as predictions, content, recommendations or decisions. The important word is "infers". The system is estimating, and an estimate can be wrong.

Three kinds of software you will meet

It helps to sort workplace tools into three groups.

The first group is rule-based automation. A person wrote the rule, and the software follows it. A spreadsheet formula that adds a column is a rule. So is an email setting that forwards every message from one client to a shared folder. The same input always gives the same output, and anyone can read the rule to see why. This is useful software, and it is not what this course means by AI.

The second group is predictive AI. These systems learned from past examples to give a score, a category or a forecast. A spam filter decides whether a message looks like junk. A bank's system flags a payment that does not fit your usual pattern. A hiring tool ranks applications. A sales tool forecasts next month's orders. The output is usually short: a label, a number, a ranking. You often cannot see the reason behind it.

The third group is generative AI. These systems produce new content: text, images, audio, video or computer code. Chat assistants belong here, along with tools that summarise documents, draft emails, write meeting notes or create pictures from a description. Most of this course is about this group, because it is the one most people now use directly, every day, with no specialist in between.

Why the difference matters

Each group fails in its own way, so each needs a different kind of care.

A rule fails when the rule is wrong or out of date. You fix it by reading the rule. A predictive system fails when the new case is unlike the examples it learned from, or when those examples carried a bias. You often cannot see the failure in a single result, only across many. A generative system fails by producing content that reads well and is wrong, incomplete or invented. That failure sits in front of you, in the text, but you will see it only if you check.

One morning, three tools

Maria works in the accounts team of a furniture wholesaler. Before ten o'clock she meets all three groups.

At nine, the accounting software sends an invoice of 6,200 euros to her manager for approval. A rule does this: every invoice above 5,000 euros needs a second signature. If Maria wants to know why it happened, she can read the rule.

At half past nine, the bank's portal marks a payment to a new supplier as unusual and holds it. No one wrote a rule about this supplier. A predictive system compared the payment with the company's history and gave it a high risk score. The score may be right, or the supplier may be new and perfectly honest. Maria has to look at the payment herself before she releases or cancels it.

At ten, she asks a chat assistant to draft a polite reminder to a customer who is forty days late. In seconds she has four tidy paragraphs. This is generative AI. The draft may be very good. It may also state a late fee her company does not charge, or strike a tone that does not suit a customer of fifteen years. Maria is the one who sends it, so Maria reads every line first.

In all three cases a machine did something useful. Only in the first could Maria see exactly why.

The common mistake

Many people think AI means the chat window. They are careful when they open an assistant and type a question, and they relax everywhere else. But the same technology now sits behind small buttons: "Summarise this thread", "Suggest a reply", "Rewrite", "Generate notes". A summary produced by a button can leave out the one sentence that mattered, in the same way a chat answer can. The care you take should follow the technology, wherever it appears.

How to spot AI in a tool

Three signs help. The tool produces sentences, images or audio that no person wrote or recorded. The tool gives a score, a ranking or a recommendation and cannot show you a rule behind it. Or the product describes a feature with words such as "smart", "assistant", "suggested" or "generated". If you are unsure, ask the person responsible for the software in your company. Knowing which of your tools are AI is the first step in using them well, and it is a question your employer should be able to answer.

Remember

AI at work is wider than chat. Rules follow instructions a person wrote. Predictive AI scores and sorts. Generative AI produces new content. All AI output is an estimate, and you remain responsible for what you do with it.

[[drill:1]]

[[drill:2]]

[[drill:3]]

[[drill:4]]

Drill 1

A support team's inbox adds the tag "Billing" to any email that contains the word "invoice". A manager set this up last year. What kind of software is doing this?

Drill 2

A recruiter's software gives every application a "fit score" from 1 to 100. Nobody at the company can say how the score is worked out. Which statement is most accurate?

Drill 3

Dimitris avoids chat assistants because he does not trust AI. Each morning he clicks "Summarise" on long email threads and acts on the summary without opening the thread. What is the problem?

Drill 4

Which output most needs to be read closely before it leaves your hands?

Continue with the full AI Literacy for the Workplace course

6 modules, 96 drills and a final exam. Pass it to become Dokimos Certified in AI Literacy for the Workplace.