How does AI work?
Kahu team·Updated 2 Oct 2026·7 min read
Most AI works in two stages. First it is trained: it is shown huge numbers of examples and adjusts itself until it can predict the right output for a given input. Then it is used: you give it new input, such as a question or a photo, and it applies the patterns it learned to produce the most likely answer.
The big idea: learning from examples
Traditional software is a recipe written by a person: if this, then that. AI is built the other way round. Instead of writing the rules, people collect examples of the right answer and let the software work out the rules for itself. That is why AI can handle messy things like handwriting, photos and everyday language, where writing every rule by hand would be impossible.
Step by step
- 1Collect dataGather examples of the task: emails marked spam or not spam, photos labelled with what is in them, or large amounts of written text.
- 2TrainThe software makes a guess for each example, checks how far off it was, and nudges millions of internal settings to do a little better next time. This repeats an enormous number of times.
- 3Keep the resultThe finished set of learned settings is called a model. It is the part that does the work from then on.
- 4Test itThe model is checked on examples it has never seen, to make sure it learned general patterns rather than memorising.
- 5Use itYou give the model new input and it produces an output: a label, a forecast, an answer or a draft.
What happens when you ask a chat assistant a question
A chat assistant runs a language model. It reads your message, then writes its reply one small piece at a time, each time choosing a likely next word based on your message and the patterns it learned. It is not searching a database of stored answers. Some assistants can also look things up on the web or read your documents first, which helps them give answers that are current and specific to you.
Why AI sometimes gets it wrong
- It predicts what is likely, not what is true, so it can produce confident errors.
- It only knows what was in its training data and what you give it.
- If the training examples were unbalanced or out of date, its answers will be too.
- Vague requests get vague, generic answers.
Where Kahu fits
Two Thursday gaps at 11:00 and 14:30. Drafted messages to 6 waitlisted customers.
Updated the hours on your site and Google profile to Sat–Sun 8–2.
One post about the new hours, using your Sunday counter photo.Frequently asked questions
Does AI keep learning while I use it?
Usually not in real time. Most models are trained, then fixed. Some tools remember your preferences or documents separately, and providers release improved models from time to time.
Where does AI get its data from?
It depends on the model. Large language models are typically trained on large collections of public text and licensed material. Your provider's privacy policy says whether your own chats are used.
Is AI just a very big search engine?
No. A search engine finds existing pages. A language model generates new text from learned patterns, which is why it can write but can also make mistakes.