What AI really is (and what it isn't) — AI Foundations

Artificial intelligence (AI) is the field of building computer systems that can do tasks we normally think need human intelligence: recognising faces, understanding speech, translating languages, spotting fraud, or writing a paragraph.

That definition is broad on purpose. AI is not one machine or one product. It's a family of techniques, and most of the AI you use every day is quiet and specialised. When your phone unlocks with your face, when your email hides spam, or when a mobile loan app decides how much you can borrow, AI is at work.

Normal software vs AI

The easiest way to understand AI is to compare it with normal software.

Normal software follows rules written by a person. A programmer writes exact instructions: "If the customer's balance is less than the price, show an error." The computer does exactly that and nothing else.

AI learns its own rules from examples. Instead of writing rules for what a spam email looks like, we show the system thousands of emails already labelled "spam" or "not spam." The system finds the patterns itself, then uses them on new emails it has never seen.

Example: recognising a matatu in a photo

Try writing rules to recognise a matatu: "It has wheels… it's long… it has graffiti…" You quickly get stuck. Buses have wheels too. Some matatus are plain. Photos are taken from different angles and in different light.

With AI, you collect thousands of photos labelled "matatu" and "not matatu" and let the system learn the visual patterns. This is why AI took over tasks like image recognition: the rules are too complicated for people to write by hand.

Narrow AI vs general AI

Every AI system that exists today is narrow AI. It's good at a specific range of tasks. A system that detects crop disease from leaf photos can't drive a car. Even chatbots that seem to know about everything are really doing one thing: predicting useful text.

Artificial general intelligence (AGI) means a system that could learn and do any intellectual task a person can. It's a research goal that companies talk about a lot, and people disagree on when, or if, it will arrive. When you see headlines about AI "thinking" or "wanting" things, remember that today's systems are powerful pattern-finders, not digital people.

Three myths to drop today

"AI understands like a human." AI finds statistical patterns. It can be very useful without understanding the world the way you do, and it can be confidently wrong.
"AI is always objective." AI learns from data made by people. If the data is unfair or incomplete, the AI will be too.
"AI is only for programmers." Most of the value of AI comes from people who understand a problem, a business or a community and know how to apply AI to it. That can be you.

A useful working definition: AI is software that learns patterns from data and uses them to make predictions or decisions on new data. Keep this sentence in mind. Almost everything in this course builds on it.

Try it. List five things you did today on your phone. For each one, decide: is it probably using fixed rules, or AI learned from data? Examples to consider: a calculator, autocorrect, your phone's photo search, an M-Pesa transaction, YouTube recommendations.

Key takeaways

AI is a family of techniques for tasks that seem to need intelligence.
Normal software follows human-written rules. AI learns rules from examples.
All AI today is narrow. General AI is still a research goal.