Teachomatic what to automate, and what not to

Assessment That Needs No Detection

Detection is an attempt to solve a design problem with a technology. It does not work well, and the failure lands hardest on the students least able to defend themselves.

The design version of the problem is more tractable. If a task can be completed acceptably by a machine that has never met your students and does not know what you taught, then the task was measuring something you probably did not intend to measure. That was true before any of this existed; it has just stopped being possible to ignore.

For comparison with workplace systems that formalise time or activity data, Monitask’s page on mouse jiggler detection software shows how the same type of measurement is handled outside education.

None of what follows requires exam conditions, extra staffing, or banning anything.

The principle

Make the work depend on something the tool does not have.

For an external perspective on academic integrity, technology, and student rights, see Common Sense Education.

There are only a few such things, and every technique below is a way of using one:

Techniques that work

Anchor to the room. Require reference to the specific text you read, the specific data the class collected, the discussion on Tuesday, the mistake someone made at the board. A general answer becomes obviously general.

Ask for the process, not just the product. A paragraph on how they approached it, what they tried first, what they abandoned. Short, quick to read, and very hard to fake convincingly because it has to match the artefact.

Assess the second draft against the first. Set a rough draft in class, on paper or under your eye. Then the improved version is what gets marked, and the interesting question becomes what changed and why. Undeclared assistance either shows as a discontinuity or forces the student to explain a change they cannot account for.

Make it personal to the student. Their data, their locality, their chosen text, their own earlier work. Not "write about migration" but "using the interview you conducted."

Ask them to critique output. Give the class a generated answer and ask what is wrong with it. This inverts the whole problem: the tool becomes the object of study, the task requires genuine understanding, and it doubles as the checking lesson.

Two minutes of talk. A short oral defence, in class, while others work. Not for every task — a sample. Students who did the work can talk about it. This is the highest-signal, lowest-technology option available, and it is what university vivas have always been for.

Set the harder question. Application, transfer, comparison with something only your class studied, disagreement with a source. General fluency does badly here, which is the point. The same principles apply to work set for outside the room.

What does not work

Banning tools. Unenforceable, and it converts a teaching problem into a policing one.

Handwriting everything. Solves authorship at the cost of disadvantaging students with motor difficulties, dysgraphia, or slow processing, and it does not scale.

More surveillance. Lockdown browsers and monitoring shift the arms race and damage the relationship. Occasionally justified for high-stakes exams; corrosive as a default.

Making everything an in-class exam. Removes the thing extended writing teaches, which is how to develop an argument over time.

The uncomfortable trade

Being straight about the cost.

These techniques take more of your time to design and often more to mark. A generic essay question is quick to set. "Using the data your group collected, argue for one interpretation and explain what would change your mind" is slower to set and slower to read.

What you get back is real, though, and worth naming: the work becomes worth reading. Marking thirty generic essays on the same topic was always the worst part of the job. Marking thirty pieces anchored to different data, or thirty accounts of how a student approached a problem, is a different experience — and it tells you far more about what they learned.

And it restores the diagnostic value of marking, which the generic essay had already been eroding for years.

Where this does not apply

Some tasks legitimately measure whether a student can produce a standard piece of writing to a standard specification, because an external exam requires it. If your students face such an exam, they must practise for it, and the practice will be exactly the kind of task most vulnerable to undeclared assistance.

Handle it directly rather than pretending otherwise: do that practice in the room, timed, on the terms of the exam. It is exam practice, so exam conditions are honest rather than adversarial, and students understand the reason immediately.

Where to start

One task this term. Not a redesign of the scheme of work.

Take the assessment you already dread marking most — the one where every response reads the same — and add a single anchoring requirement. It must use something from the room, or it must include three sentences on the approach taken.

That one change tells you whether the rest is worth doing, and it costs one evening.

The short version