By Finn Toompuu · Founder, Qualidact · Published
Assessment has always rested on a quiet assumption: that the work a student hands in is a window into what they know. Write an essay, solve the problem set — the product stood in for the person. Generative AI has cracked that assumption open. When a system can draft a competent essay or solve a problem in seconds, the product no longer reliably tells us anything about who submitted it.
The instinctive response has been to treat this as a cheating problem — detection software, honor codes, bans. But detection tools produce false positives, penalize non-native English writers, and are easily evaded. More fundamentally, the cheating framing misses the deeper issue. If a task can be completed convincingly by a machine, we should ask what that task was really measuring in the first place.
The uncomfortable audit
Much of what schools have assessed was, honestly, the ability to produce a certain artifact — an essay, a summary, a standard analysis. These were proxies, used because they were scalable ways to infer thinking. AI has exposed how thin some of those proxies were: a polished summary might demonstrate comprehension, or only fluency and formatting. This is an invitation to audit. For every assessment: what capability am I trying to certify, and does this task still provide evidence of it?
What still matters
Several capacities remain stubbornly human. The first is judgment — evaluating whether an answer is right, biased, or incomplete. AI outputs are fluent even when wrong, making this skill more important, not less. The second is understanding that survives novelty: real comprehension lets someone adapt and explain why a method works, where memorized or generated answers collapse under a shifted question. The third is framing problems — deciding what's worth asking. AI answers questions; people still decide which ones matter. Finally, there is accountability: someone must stand behind a decision and bear its consequences.
Redesigning assessment
Oral examinations are gaining renewed attention because they reveal understanding in real time and resist outsourcing. Process-based assessment — drafts, notes, reflections on reasoning — shifts attention from product to thinking. In-class, low-tech work establishes a baseline of unaided ability. Equally important is assessing AI-assisted work openly: evaluating how well someone directs, critiques, and improves AI output is itself a legitimate skill. None of this is free — oral exams and process portfolios demand more time, and equity concerns arise when tool access differs. Institutions must invest in assessment design rather than treat it as an afterthought.
A better question
AI forces a clarifying question education should have asked all along: what do we want people to be able to do, understand, and take responsibility for? The goal isn't protecting old tasks from new tools — it's looking through the artifact to the person behind it. AI hasn't made that harder to see; it has made it impossible to keep ignoring that we were ever only guessing.