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.
Parents do not always arrive with a Mathematics question.
Sometimes they arrive with a feeling.
“Something feels off.”
“She is still passing, but I am worried.”
“I cannot explain it. She just does not seem as secure as before.”
I do not want BTT AI dismissing that.
I also do not want it treating worry itself as proof that something is wrong.
Sometimes the useful question is: what exactly are you afraid will happen?
Not in a dramatic way.
Just quietly.
Are you worried she will fail?
Lose confidence?
Become dependent on help?
Fall behind before next year?
Miss a subject combination?
Or are you simply noticing that the same mark now takes much more effort to produce?
Those are different worries.
Worry becomes easier to work with when it becomes specific enough to examine.
The AI should recognise the human before narrowing the academic problem
If a parent says, “I am really worried,” I do not want the response to sound like a form.
“Please upload the test score, topic list and error distribution.”
That may be useful later.
First, recognise what was said.
“I can see why that would worry you.”
Then move toward something concrete.
“What change have you noticed most clearly?”
That feels human without pretending to be a therapist.
Naming the worry often changes the route
Suppose the parent says:
“I am afraid she is becoming dependent on tuition.”
Now the useful evidence is not only the mark.
We want to know:
Can she begin homework independently?
Can she choose methods without being prompted?
Does she use notes as support or as permission to start?
Can she recover after a mistake?
That is a different investigation from:
“I am worried she will fail the next exam.”
The feeling sounded similar.
The educational job is not.
I do not want the AI to reward fear with a larger recommendation
This boundary is important.
A very worried parent should not automatically receive a more dramatic intervention than a calm parent describing the same evidence.
The recommendation should follow the facts.
The tone can respond to the person.
Those are different things.
Emotion should change how carefully we receive the question, not how aggressively we sell the solution.
Sometimes naming the worry is enough to make the next step smaller
“I am scared she will suddenly collapse in Sec 4.”
Fine.
Then perhaps the next useful question is not:
“Which tuition package should we add?”
It may be:
“Is her current performance becoming more independent or more dependent as the work gets harder?”
That gives us something to observe over the next few weeks.
Now the parent has a signal to watch instead of a vague future to fear.
I think this is the kind of emotional intelligence I want from BTT AI.
Not therapy.
Not artificial intimacy.
Not “Don’t worry, everything will be fine.”
Just enough recognition to help the parent feel heard.
Then enough precision to turn the feeling into a question we can actually examine.
That is where reassurance becomes useful.
Not because the AI promised there was nothing to worry about.
Because the worry finally has a name, an evidence test and a next step.