AI design essay: This article sets out the standards we would want educational AI to meet. The examples illustrate proposed behaviour; they do not establish that a deployed BTT AI currently performs these functions. Mathematics advice should be checked against the learner’s actual work and the teacher’s evidence.
There is a future version of educational AI that worries me slightly.
The student becomes excellent at asking the AI.
She knows how to phrase the prompt.
She knows when to ask for a hint.
She knows how to get a cleaner explanation, a worked example, another method and a summary.
Everything looks sophisticated.
And yet, when the AI disappears, so does half the performance.
I do not want that.
I do not want your child to become impressive at operating the support.
I want the support to make her more capable without it.
There is a difference between using help well and needing help well
A student may ask excellent questions:
“Give me one hint only.”
“Do not solve it. Ask me what I should notice.”
“Show me where my first wrong line is.”
I like those requests.
They are much better than “Give me the answer.”
But even good use of help can become a habit of dependence if the learner never has to operate without the help.
So after a while, I want the AI to step back.
Try this one without me.
Explain your choice before asking.
Check your own answer first.
Come back only if the same difficulty remains.
The quality of the AI should eventually appear as quality in the learner, not as permanent activity around the learner.
I would watch what the AI is carrying
Is it choosing the method?
Breaking the problem into steps?
Checking every line?
Reminding the student what formula belongs?
Explaining every error immediately?
Any one of those may be useful for a while.
The question is whether responsibility is moving.
At first the AI may carry a lot.
Later, less.
Eventually, perhaps almost nothing unless the learner reaches a genuinely new edge.
The student should become better at asking herself the questions
This is where I think AI can be genuinely educational.
At first the system asks:
“What do you already know?”
“What are you trying to find?”
“Why did you choose that method?”
“Does your answer make sense?”
Later, I want those questions to migrate inward.
The student begins asking them before the AI appears.
That is transfer.
The prompt has become part of the learner’s own thinking.
And yes, there will still be times when the AI is useful
I am not arguing for artificial hardship.
If the student genuinely does not understand a new concept, explain it.
If she has tried three routes and cannot locate the error, help her narrow it.
If she wants to compare two methods after solving, that can be excellent use of AI.
The goal is not to make the AI scarce.
The goal is to make the help proportionate.
If, six months from now, your child can produce beautiful prompts but still cannot begin an unfamiliar Mathematics question without the AI, I would not call that success.
If she uses the AI less often because she now recognises more, checks more, explains more and recovers more by herself, I would.
The product should not be an expert user of BTT AI.
The product should be a more capable student.