How we use AI — and how we ask students to.
AI is changing how mathematics is taught and how students try to learn it. This page sets out our department's position so parents and students know what to expect, rather than guessing. It is not a list of bans — it's a frame for thinking about when AI helps and when it gets in the way of the actual learning.
Our underlying view of learning
Daniel Willingham puts it bluntly: "Learning is a change in long-term memory; if nothing has changed in long-term memory, nothing has been learned." That sentence is the lens we use for every decision about AI. The question is never "did the student get the right answer?" It is "did the student do the work that produces the change in long-term memory?"
A second principle, also from cognitive science: critical thinking is not a generic skill that exists independent of knowledge. You cannot think critically about a topic you know nothing about. Building the knowledge base is the slow part. AI can produce answers far faster than students can build the understanding that lets them evaluate those answers — and that asymmetry is the heart of the problem.
So we aim for understanding, not just correct answers. The trouble is that students often, reasonably, want to short-cut. A small example: when we introduce the special angles (30°, 45°, 60°) at the start of trigonometry, we teach them through the symmetries of the unit circle and the geometry of the equilateral and right-isosceles triangles, so a student understands why cos 30° = √3/2 and can reconstruct the value at any related angle (60°, 120°, 150°, 330°…). Some students just memorise the table of values instead. It works in the short run. But the student who understands the construction can derive any value they need, including ones the table doesn't list — and remembers it longer because the memory is anchored in something. AI is a much faster table.
How teachers use AI
We use AI to enhance the lesson and build resources that would not have been practical to make before. Variations of a problem at three different difficulties for the same class. A misconception sweep before a topic so we plan around what students will trip on. Worked-example sets pitched at a specific student's gap rather than the average. Drafting starting points for written feedback that we then edit. AI does the bulky generation; we keep the judgement, the structure, and the choice of what serves the class.
The biggest practical gain is personalisation. With AI we can produce resources tailored to an individual student or a small group — a recap sheet using only the topics one student missed last week, an extension stream for a student who has finished the unit, a French-language version of a worksheet for a student who would otherwise stall on the English. None of that would scale by hand. With AI it does.
How students may use AI
Different year groups, different appropriate uses:
- Y7–Y9For explanations of concepts you're stuck on (after you've genuinely tried). Not for completing set problems — those are how you build the long-term memory the AI can't build for you.
- Y10–Y11Same as above, plus checking your own work after attempting it yourself and being able to point at what you got wrong and why. If AI is involved in a written task, say so.
- DPComputational tools (Wolfram Alpha, Desmos, GeoGebra) are normal and expected. AI assistants are useful for explanation and exploration; if you use one in coursework or the IA, disclose it. Examiners are increasingly explicit about this — get the habit early.
But don't rely on it. AI almost never follows the journey your teacher is curating. We introduce most ideas first from a concrete, conceptual angle and only later move to the standard algebraic shorthand. AI tends to jump straight to the formal version, which means you get a correct answer using machinery you have not built yet — and the conceptual scaffolding gets skipped.
Concrete example: in sequences, if the 2nd term is 5 and the 6th term is 17, the conceptual way to find the common difference is to count the jumps. There are four jumps between the 2nd and the 6th term, so the total change of 17 − 5 = 12 is split across 4 equal jumps: each jump is 3. That is the understanding we want students to build. Ask AI and it will write , set up two equations, and solve — the formula you will meet later in the course. Both reach 3. Only one teaches you what a common difference actually is.
Where AI is not appropriate
Anywhere the work is about you doing the thinking. That includes: homework problems whose purpose is to build fluency; in-class formative tasks; the IA Exploration; exam-style practice you haven't attempted yet; anything you submit as your own.
If we ask "show your working," the working is the point — not decoration. AI can produce a finished derivation without any change in your long-term memory. That's a problem because next week we'll build on it, and you won't have what we're building on.
Tools we trust (and what they're for)
Wolfram Alpha
Visit ↗Computational engine. Best for checking algebraic manipulation, plotting, simplifying, and verifying numerical answers. Not designed to teach — designed to compute.
Desmos
Visit ↗Free graphing calculator. We use it across MYP and DP for visualising functions, transformations, calculus, and statistics. The strongest single tool for "play with the maths to see what it does."
GeoGebra
Visit ↗Free dynamic geometry / algebra / calculus environment. Particularly good for geometry investigations and constructions.
General-purpose AI assistants (ChatGPT, Claude, Gemini, etc.)
Useful as patient tutors for explanation. Less reliable as answer-machines — they confidently produce wrong working. Always sanity-check against a textbook or a teacher. Disclose when you use one in submitted work.