When people hear the phrase artificial intelligence, many minds jump straight to a scene from a film: a machine that thinks, wants, and decides to get rid of us.
And when you tell them that artificial intelligence is what sorted their inbox this morning, there is a small feeling of disappointment.
The truth is that the phrase covers two completely different things: one is in your phone, the other does not exist yet.
What is narrow AI?
Narrow AI is a system that is good at one task, or a defined set of tasks, and cannot step outside them.
The system that recognises your face does not know how to translate a sentence. The one that drives a car does not know how to diagnose an illness. And even the model that writes you a full article cannot ride a bicycle.
Every artificial intelligence that exists today, without exception, is narrow AI. Including the most powerful and most impressive models.
And what is general AI?
General AI is a hypothetical system able to learn any intellectual task a human can, move between fields, and apply what it learned in one context to a completely new one.
Notice the word hypothetical. This type does not exist. Not in a lab, not in a company, not in a secret version.
And the disagreement among researchers is not only about when it might appear, but about whether the current path leads there at all.
Why the confusion?
Because today’s systems look general when they are not.
A single language model writes poetry, explains a law, fixes code, and translates. That looks like enormous range.
But it is really one task repeating: predicting the next piece of text. Poetry, law and code are all text. The model did not learn four fields. It learned one field that appears in four forms.
This is why you see the same model explain a complex theory beautifully, then fail at simple arithmetic. If it genuinely understood, that strange gap would not exist.
Why does this distinction matter to you?
Because it sets what you can actually expect.
Once you know the system is narrow, you stop waiting for judgment, understanding, or responsibility from it. And you start treating it as it is: very powerful inside its range, and completely helpless outside it.
This changes practical decisions. You will not hand it a call that needs human context. And you will not assume it will notice that your question itself was wrong.
In return, the distinction protects you from generalised fear. The debate about AI risk today is not about a machine that woke up. It is about narrow systems being used in large decisions without enough oversight.
In short
Narrow AI is a daily reality you have lived with for years. General AI is a theoretical idea whose feasibility is still disputed.
And when the two blur together in people’s minds, two things follow: expectations larger than the tools can meet, and fears aimed at the wrong place.
The danger today is not a machine that wants something. It is a tool that wants nothing, handed decisions that needed someone to answer for them.