Three robots, three very different minds
In the scene above there are three robots. They look similar, but their minds could not be more different. Let’s meet them one at a time.
Robot 1: Narrow AI — the one-trick genius
The small blue robot is narrow AI. It is amazing at exactly one thing.
Maybe it plays chess better than any human alive. Maybe it can spot a cat in a photo in a thousandth of a second. Maybe it suggests the next video you might like. Whatever its one thing is, it is superb at it.
But ask it to do anything else and it is completely lost. The chess robot cannot play draughts. The cat-spotting robot has no idea what a dog is unless someone teaches it separately. The video-suggesting robot cannot tell you a joke.
Almost every AI you have ever used is narrow AI. The voice assistant in a phone, the thing that unlocks a phone with a face, the tool that translates a menu on holiday — all narrow.
What about chatbots that seem to talk about anything? They are broader than old narrow AI, and some people call them “jagged”: very good at some things, surprisingly bad at others, and unable to learn new skills on their own after training. They are not yet the next robot.
Robot 2: AGI — the all-rounder
The green robot is AGI, which stands for artificial general intelligence. The important word is general.
An AGI could learn almost anything a person can learn. Show it a new board game and it would work out how to play. Give it a job it has never seen and it would figure it out, the way a new employee does. Ask it to explain its thinking and it could.
In other words, AGI would be roughly as capable as a human — not better than the best humans, just able to move between tasks the way we do. On the intelligence staircase, AGI is a machine reaching the human step.
Nobody has built an AGI. Some researchers think we are close. Others think important pieces are still missing. You can see their arguments on Is superintelligence here yet?
Robot 3: Superintelligence (ASI) — beyond everyone
The giant glowing robot is superintelligence, sometimes written ASI for artificial superintelligence.
ASI would do everything AGI can do — and be far, far better at it than the best humans. Better at science than the best scientist. Better at planning than the best planner. Better at understanding people than the wisest person you know.
It would also have advantages that come from being software. It could think thousands of times faster than a brain. It could be copied, so millions of versions could work on different problems at the same time. It would never get tired or forget.
That is why the third robot is drawn so big. It isn’t stronger — it’s that its mind is on a completely different scale. And, like AGI, it does not exist. It is an idea that people are trying to understand before it arrives, if it ever does.
The comparison table
| Narrow AI | AGI | Superintelligence (ASI) | |
|---|---|---|---|
| What it can do | One job, extremely well | Almost any job a person can learn | Almost everything, far better than the best humans |
| Learns new skills by itself? | No | Yes, like a person | Yes, and very fast |
| How it compares to people | Better at its one thing, useless at the rest | About the same as a person | Far beyond every person combined |
| Does it exist? | Yes — everywhere | No (still being worked on) | No (an idea) |
| Example | A chess program, a face-unlock, a spam filter | None yet | None yet |
| On the staircase | Off to the side — a tall, thin spike | The human step | The glowing steps in the clouds |
Why the difference matters
People mix these up all the time, and it causes confusion. When someone says “AI is going to change everything”, they might mean narrow AI (already true), AGI (maybe soon, maybe not), or ASI (a much bigger deal). Those are three very different claims.
Keeping them apart helps you think clearly:
- Narrow AI is here. The questions are about how we use it well.
- AGI is a goal. The questions are about whether and when it will be reached.
- Superintelligence is a possibility. The questions are about how to make sure it would be safe and helpful — before it exists, not after.
Next time you hear a headline about AI, ask yourself: which robot are they talking about?