Academic Integrity: Our Position

Using a solver to understand a method is learning. It is not different in kind from a worked example in a textbook, a tutor at a whiteboard, or the answers in the back of the book. Submitting its output as your own work is not learning, and it is the thing every academic misconduct policy in the world is about.

The line is not drawn by how much help you got. It is drawn by whose understanding the submitted work represents. Two students can spend the same twenty minutes with the same solution open. One reads it, finds the step where their own attempt went wrong, closes the tab, and rewrites the problem from scratch. The other transcribes it. The first one owns the work; the second one does not.

One practical test settles most cases. Shut the laptop and ask whether you could produce that solution again tomorrow, unaided, on a blank page. If yes, the understanding is yours and the write-up should be too. If no, what you are about to submit is a transcript.

Four concrete cases, because principles are easy to agree with and hard to apply. You have finished a problem set and want to check your answers: fine in most courses, and no different from the back of the book — unless it is a take-home exam, where checking is still outside help. You cannot work out what a question is asking and use the solver to unpick the wording: fine, and particularly reasonable if you are studying in a second language. You are stuck on one step in the middle of a derivation, read that step, and finish the rest yourself: fine, and it is the use this product is designed around. You have not started, the deadline is in an hour, and the plan is to send the question and copy out whatever comes back: that is the case everyone already recognises and nobody wants to name.

Our opinion is not a permission slip. Policies vary between institutions, between departments in the same institution, and between assignments on the same course — the same tool can be encouraged on a problem set and expressly banned on a take-home exam two weeks later.

Look in three places before you use any AI help on marked work: the assignment brief, the module or course handbook, and the academic misconduct policy. If all three are silent, ask whoever is marking it. That question has never got a student into trouble, and the answer takes an email.

Where a policy allows AI with disclosure, disclose. A single sentence saying what you used and for which step costs nothing when it is permitted and converts a risk into a non-event.

The argument against copying that lands with students is not moral, it is practical, and it is about timing. Copying has no immediate cost. The homework gets done, the marks come back fine, and the feedback loop that would tell you something is wrong never fires. The bill arrives in a hall with a paper in front of you and no tab to open, and by then there is no time left to learn the term you skipped.

There is a second cost that shows up sooner. Marks in most mathematics assessments are attached to method, not to the final number, so a submission that reproduces someone else’s route often loses marks anyway — for skipping the justification the marker was looking for, or for using a technique the course has not covered. How markers actually allocate method marks breaks that down line by line.

The thing that does work is boring and well evidenced: attempt first, get unstuck, then reproduce the solution from memory a day later. What the research says about studying for a math test covers retrieval practice and spacing, which are the two ideas doing most of the work.

We cannot stop anyone copying, and we are not going to pretend a product decision can. What we can do is make understanding cheaper than copying. Solutions come as numbered steps with a justification attached to each move, rather than a bare answer, because a justification is something you can argue with. Follow-up questions are unlimited and free on every plan, so asking why step four is valid never costs you a solution from your daily quota. Every answer carries a line reminding you that it can be wrong.

Using AI for homework without cheating goes further into the grey cases — checking a finished answer, translating a problem you cannot parse, working in a second language — with a self-test for the ones that are genuinely unclear. If you would rather build the method than borrow it, the concept guides derive each technique instead of asserting it.