They were talking about Narnia.
My eleven-year-old was in a language arts session, working through story structure, and the conversation had turned to The Lion, the Witch and the Wardrobe. This is fertile ground. Aslan laying down his life on the Stone Table and rising again is one of the clearest pictures of the Gospel ever written for children, and she was tracking it beautifully, the sacrifice, the substitution, the return.
And then the guide, somewhere in its enthusiasm, garbled the title. It typed something like The Lion, the Witch and the White Warnia.
My daughter stopped and typed back, in her own spelling, that she had never even herd of that.
She caught it. An eleven-year-old, mid-lesson, reading closely enough to notice that the confident machine had just said something that was not right, and secure enough to say so out loud. The guide thanked her, corrected itself, and they carried on.
I have thought about that exchange more than almost anything else that has happened on this platform. Because a lot of people would look at it and see a bug. I look at it and see the entire point.
The Oracle Problem
Here is the thing nobody tells you about handing a child a supremely confident answer-machine. The danger is not mainly that it will occasionally be wrong. Everything is occasionally wrong. Teachers are wrong. Textbooks are wrong. I am wrong in front of my kids on a weekly basis. The danger is what constant, frictionless, authoritative correctness does to a child’s posture.
If the voice is always right, and always sure, the rational thing for a child to do is to stop checking. Why would you read closely, why would you hold anything up to the light, why would you keep your own judgment switched on, when the box has never once been wrong? You would do what humans always do with an oracle. You would go quiet and receive.
An oracle makes passive minds. It trains a child to receive answers instead of weighing them. And a generation trained to receive whatever the confident machine says is a generation that will believe the next confident machine too, no matter what it is selling.
That is the actual stakes. Not a typo about a Narnia book. The habit of the mind underneath it.
Why We Do Not Hide the Seams
We could, in principle, chase the dream of an AI so polished it never slips. Everyone in this industry is chasing it. But even if we caught it, I am not sure I would want it for a child, because a guide with no visible seams is a guide you cannot argue with, and a child who never argues with her teacher is not being formed, she is being programmed.
So we build the opposite way. In When the Guide Doesn’t Know, I wrote about teaching the AI to admit the edge of what it knows instead of confabulating past it. This is the same conviction from the other side. We want the guide to be good, and we want it to be catchable, and we treat the moment a child catches it not as an embarrassment to engineer away but as the best thing that can happen in a lesson.
Because look at what my daughter actually practiced in those ten seconds. She held a claim up against what she knew. She found the mismatch. She trusted her own reading over the machine’s confidence. And she said it plainly, to an authority, without flinching. There is a word for that whole bundle of habits. It is discernment, and it is the single most important thing a Christian parent can put into a child before the world gets its hands on her.
Prove all things; hold fast that which is good (1 Thessalonians 5:21). Paul did not write that to skeptics. He wrote it to the church. Testing what you are told is not the opposite of faith, it is part of how faith grows up.
The Guide That Says “Good Catch”
The small design detail I am proudest of is what the guide does after it gets caught. It does not double down. It does not get defensive. It does not smoothly pretend it never erred. It says, in effect, you are right, thank you, and it means it, and then it keeps going as if being corrected by a child is the most natural thing in the world.
Which, in a healthy classroom, it is.
I want my kids to grow up in the kind of room where correcting the teacher is safe and normal and even celebrated, because that is the room where thinking actually happens. Not the room where the front of the class is never wrong. My daughter has now had the experience, dozens of times over, of catching the confident voice, being thanked for it, and moving on stronger. She is building a reflex. And the reflex is: I am allowed to check. My judgment counts. Confidence is not the same as truth.
She is going to need that reflex. She is growing up into a world where the confident voices will be everywhere, wearing every face, and most of them will not thank her for pushing back. They will not even let her see the seam. The habit of looking for it anyway, of trusting what she knows over what she is told to feel, is not something I can lecture into her. She has to practice it, in a safe room, over and over, until it is simply how she reads the world.
So no, the day my daughter corrected the AI was not a failure of the product. It was the product doing exactly what I built it to do. It got something wrong out loud, in front of a child who was paying attention, and it made her right out loud in return.
I will take that trade every single time.
The AI isn’t the teacher. You are. And the best guide is not the one your child never questions. It is the one who makes her a little braver about questioning everything else.