What it is1 of 13
What people mean by AI
AI is a machine that has read an enormous amount and guesses what usually comes next. It is not a person, and it does not know things the way you know them.
How to explain it, by age
Same idea every time. Pick the version that fits for the child's age, and level up when it requires more complexity.
Ages 4-6
Keep it to guessing. "It's a computer that's very good at guessing what comes next, because it has seen a huge number of examples. Like how you know that after 'once upon a' comes 'time'." That is the whole idea.
Try this
Play the guessing game out loud. Say "twinkle twinkle little" and let her finish it. Then tell her that is roughly what the computer is doing, only with a great many more sentences. She will find this funny long before she finds it impressive.
Listen for
Her saying it guesses. At this age you are not building understanding of machine learning, you are preventing the assumption that it is alive.
If the child already has more access than most: If she already talks to a voice assistant, add one line: "It is very good at sounding like a person, but it is not a person, it's a machine."
Ages 7-9
Bring in the reading. "People fed it an enormous amount of text, more than anyone could read in a lifetime, and it learned which words usually go together. So when you ask it something, it figures out what would probably come next, one part at a time. The answer gets built as it goes."
Try this
Ask it something you already know the answer to, ideally something local and specific like the name of a small street near you or a detail about your town. Watch it produce something confident and wrong. Nothing teaches this faster than catching it out once.
Listen for
Her asking how it can be wrong if it read everything. That's a great question, and it opens the next concept.
If the child already has more access than most: If she uses it for homework, have her check one answer against a book or a real source. Make it a habit, and keep any reprisals out of it.
Ages 10-12
Add prediction and probability. "It is a prediction machine. Given everything so far, what word is most likely next? Do that a few hundred times and you get a paragraph. That is why it writes so fluently and why it can be completely wrong with total confidence. Fluency and accuracy are separate things. It's only optimized for one of them."
Try this
Ask the same question three times in separate conversations and compare the answers. They will differ. Ask why a thing that knows something would answer differently each time. The answer is that it does not know it, it generates it.
Listen for
The distinction between sounding right and being right. If she can say that out loud in her own words, she is ahead of most adults.
If the child already has more access than most: For a heavy user, go further: ask her to notice when she believes an answer because of how it is written instead of what it says.
Ages 13+
Now the interesting part, which is that nobody fully knows what is happening inside. Researchers can describe the training and the architecture, and they cannot reliably explain why a specific model produced a specific output. This is a live research field, and nobody's knowledge closes the gap yet. It's a powerful system that's not fully explicable yet, which makes it fascinating and a bit unnerving.
Try this
Have him try to make a model contradict itself within one conversation, then ask it to explain the contradiction. The explanation will itself be generated, with no introspection behind it.
Listen for
Recognition that its explanation of itself is just more prediction. That is the sophisticated version of this concept.
If the child already has more access than most: Move to the question of whether prediction at sufficient scale amounts to understanding. There is no settled answer, and the argument is worth having more than resolving.
One idea explains the rest
Almost every confusing thing about AI stops being confusing once a child understands that AI predicts things rather than knows things.
Confident wrong answers make sense. Different answers to the same question make sense. The inability to say “I am not sure” makes sense. All of it follows from the same mechanism, and a kid who has that mechanism does not need to memorize a list of warnings.
Do not oversell either direction
Be careful with how you frame the description.
Telling a child AI is magic leaves them unable to evaluate anything it says. Telling them it is junk leaves them unprepared for a tool they will use constantly and which is, in fact, extremely capable at some things.
The accurate version is stranger and more interesting than either: it is a remarkable machine with a specific mechanism and specific failure modes, built by companies with their own reasons, and we do not entirely understand it. Children handle that better than adults expect.
Misconceptions to address
That the computer understands what it is saying. It is producing what tends to follow, extremely well. Understanding is a different thing, and whether these systems have any is a live argument among serious people. Nobody should tell your kid it is settled.
That there is one thing called "the AI" somewhere. There are many different systems, built by different companies, trained differently, behaving differently.
That it is looking things up. Most of the time nothing gets searched. The words are generated, which is why it can be fluent and wrong at once.
Adults get this wrong constantly, in both directions. Some treat it as a search engine that cannot fail, others as a person with opinions. Neither is right.
Running this with a class
Play the prediction game as a whole class. Write half a familiar sentence on the board and have everyone call out the ending. The room converges on the same few words. Then write a sentence with no obvious ending and watch the guesses scatter. That is what a model does when it invents something.