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.
Imagine two months with the same Mathematics mark.
72 in June.
72 in August.
If you only look at the number, nothing happened.
But suppose June’s 72 needed:
- three reminders to start revision;
- notes open beside every question;
- frequent hints;
- a parent checking each answer;
- a tutor identifying the topic before the student chose a method.
And August’s 72 needed almost none of that.
I would not call those two performances equal.
The mark is flat.
The learner is not.
This is the kind of progress I want BTT AI to notice
Not only:
Did the score rise?
Also:
How much help was required to produce the work?
How often did the student need the first step?
Could she choose the method herself?
Could she detect an unreasonable answer without being told?
Could she recover after a wrong turn?
Did she come back to the AI for every uncertainty, or only for the genuinely difficult ones?
A stable result produced with less external support can be real progress.
Less help should not become a vanity metric
I want to be careful here.
The goal is not to make the student struggle alone so that we can congratulate ourselves on independence.
Some new topics genuinely require teaching.
Some difficult questions deserve discussion.
A student who asks for a precise piece of help at the right moment may be more independent than one who refuses help and repeats the same mistake for forty minutes.
So I am not measuring independence by silence.
I am looking at whether the learner increasingly owns the ordinary decisions she is now capable of making.
The type of help should change too
At the beginning, perhaps the AI says:
“Use simultaneous equations here.”
Later:
“What two unknowns are linked by the information?”
Later still:
“What structure do you recognise?”
Then eventually:
nothing.
The student has already started.
The support has not merely become smaller.
It has moved upward: from giving the method, to prompting recognition, to observing independence.
This gives parents a calmer way to read a plateau
Parents understandably become concerned when marks stop rising.
Sometimes the plateau is genuinely a sign that something needs attention.
But sometimes the student is consolidating underneath the same score.
The questions are harder.
The support is lower.
The mistakes are narrower.
The recovery is faster.
Those details do not guarantee the mark will rise next month.
They do tell us that “no progress” may be too crude a description.
If BTT AI remembers anything about a learner over time, I want it to remember this kind of movement.
Last month you needed the first step.
This month you usually find it yourself.
Last month every wrong answer became a request for help.
This month you check before asking.
That is not decorative progress.
That is responsibility moving into the learner.