What artificial intelligence actually is

AI is a category, not a single thing

An email spam filter, a recommendation system and a conversational assistant can all be called AI, even though they work differently and solve different problems. The useful first question is therefore not simply “Is this AI?” but “What kind of system is this, and what was it built to do?”

Most current AI systems are specialised. They can be remarkably capable within their area while lacking the flexible understanding people naturally bring from one situation to another.

How does a system learn?

Many modern systems learn patterns from examples rather than following only a list of hand-written rules. During training, a model adjusts internal numerical relationships so that it becomes better at predicting an appropriate output from an input.

Learning patterns is powerful, but patterns are not guarantees. Training data can be incomplete, biased or outdated, and a convincing answer can still be incorrect.

What makes generative AI different?

Generative AI creates new outputs—such as text, images, audio or code—based on patterns learned during training. A language model works by predicting likely continuations of language. With enough scale and careful training, that prediction can support explanation, summarisation, planning and conversation.

What can it actually know?

It is safer to think in terms of access and reliability. Does the tool have current information? Was it given the relevant context? Can its answer be checked? A model may hold useful learned patterns, but it does not automatically know your circumstances, the latest event or whether its own response is true.

Good use begins when capability and uncertainty are considered together.

Apply: try a small, everyday task

Use a task where you can judge the result yourself. For example, ask an AI assistant to turn a made-up list of jobs into a plan. Copy this prompt into the assistant you use:

I have 45 minutes to tidy my desk, water my plants and organise a shopping list. Suggest a simple plan that leaves five minutes spare. Explain your timing and tell me what you have assumed.

Check that the times add up and that the plan fits your situation. Then try a follow-up: “I only have 20 minutes. What should I prioritise, and why?” Compare the two plans and choose what actually helps.

You can practise with invented details. You do not need to share personal documents or private information for this exercise.

Question: check before you rely on it

A clear answer is a starting point for checking. Use these questions as you read:

• What is the claim? Separate checkable facts from suggestions and opinions.
• What supports it? Open any cited source. Check that it exists and supports the specific claim.
• What is missing? Look for assumptions, missing context and information that may be out of date.
• Does it make sense? Recalculate totals and compare the answer with what you already know.
• What will I use? Decide which parts to keep, change or reject. Asking the same assistant again is not an independent check.

For the planning example above, add up the minutes, check the five-minute buffer and decide whether each job can realistically fit. A useful-looking plan still needs your judgement.