Skip to content

Education · 7 min read

Adaptive assessment without pretending the model is infallible

Adaptive assessment is useful precisely because it is uncertain: it narrows a question rather than settling it. Presented as a verdict, the same mechanism becomes indefensible.

What the model is good at

Choosing the next question, spotting the competency where a cohort clusters below target, and drafting variants of an item an instructor then edits. All three save hours and none of them issue a judgement about a person.

A recommendation an instructor cannot overrule is not decision support. It is an unaccountable decision.

Where it has to stop

Certification, progression and anything that appears on a transcript belong to a human evaluator. The platform's job is to present the evidence and the confidence, then get out of the way.

In practice that means every AI-assisted suggestion carries a reviewer, a timestamp and a way to reject it without leaving the screen.

Explain the gap, not just the score

A learner told they scored 62 % learns nothing. A learner shown which competency is below target, and which two modules address it, has a next action.

The competency model, not the score, is the artefact worth investing in.

Related solutionLearning & Skills →