What AI is, and what it is not
"Artificial intelligence" is used for so many things that it can mean almost nothing. This lesson gives you a working definition, the main kinds of AI you will meet, and a few myths worth dropping.
A working definition
AI covers many techniques, but most of the AI systems you will meet today are software that learns patterns from examples and then uses those patterns to produce an output: a prediction, a classification, a recommendation, a piece of text or an image.
That is different from ordinary software, where a programmer writes every rule by hand ("if the invoice is over 30 days old, send a reminder"). With these systems, people supply examples and a goal, and a training process adjusts the system until its outputs match the examples well. The rules are learned rather than written.
The key idea: most of today's AI is pattern-matching learned from data. It is only as good as the examples it learned from and the way it is checked.
The kinds of AI you are most likely to meet
| Kind | What it does | Everyday example |
|---|---|---|
| Predictive models | Estimate a number or choose a category from structured data, such as rows and columns in a spreadsheet | Which customers are likely to cancel; next month's demand |
| Language models (LLMs) | Generate and transform text by predicting likely next words | Chat assistants, drafting emails, summarising documents |
| Image, audio and video models | Recognise or generate images, speech and video | Photo search, voice dictation, image generators |
| AI agents | Use a model to decide on a sequence of actions, often by calling other tools | An assistant that searches, books or updates records for you |
Many products combine several of these. A customer-service assistant might use a language model to read a message, a predictive model to estimate urgency, and an agent to look up the order.
What AI is not
It is not a person
Today's AI systems should not be treated as having human understanding, judgment or accountability. A language model can write fluently about a topic without any way of checking whether what it writes is true. Fluency is not the same as knowledge, and responsibility for a decision stays with the people who make it.
It is not automatically objective
Because these systems learn from examples, they inherit whatever is in them, including gaps, errors and historical bias. A model trained mostly on one group of customers may work poorly for another.
It is not a guarantee
Every AI system makes mistakes. The useful questions are how often, what kind, and whether anyone has measured it. A system nobody has tested on fresh data has an unknown error rate, however impressive its demonstration.
It is not one thing
"Is AI reliable?" has no single answer. A well-tested model that predicts equipment failures from sensor readings and a chatbot answering legal questions are very different tools with very different risks. Ask which AI, doing what, checked how.
Three questions to ask about any AI system
- What did it learn from? Whose data, how much, and how recent?
- How was it checked? On data it had never seen, and against what standard?
- What happens when it is wrong? Who notices, and how costly is the mistake?
The rest of this course is about answering those three questions well.