How this editorial can be challenged
What should a university actually certify when it says a graduate can work competently with AI?
Assessment rewards shape habits. If courses reward only completion, students can learn to outsource judgment; if they require traceable evidence, explicit permission and appropriate escalation, useful automation becomes compatible with retained human agency.
Too many checks could turn practical AI education into an obstacle course, consuming time that learners need for experimentation and useful work.
The answer is proportionate friction: broad freedom in harmless practice environments, stronger checks when actions affect other people, and credit for recognizing when no escalation is needed.
Today's educational sources establish a scheduled program and a competency framework, not causal evidence that our proposed assessment improves learning or employment. The water study supports a bounded research example, not validation of a curriculum.
Controlled classroom comparisons showing no improvement in error recognition, transfer or appropriate escalation, or showing prohibitive accessibility and staffing costs, would weaken this proposal.
The moment after the answer
Imagine a student looking at a beautifully written answer. The argument is clear. The references look respectable. The task appears finished. Then the student notices that one reference does not support the claim. Nothing dramatic happens: no alarm, no broken screen, no obvious instruction to stop. The consequential decision is whether to submit it anyway.
This is a hypothetical classroom, not an anecdote from a study. It is also the scene I would put near the centre of practical AI education. The skill we need is not permanent distrust. It is recognizing the moment when an apparently successful output has not yet earned permission to become a decision.
A university is more than a software classroom
New York Tech's scheduled AI Week offers a useful institutional starting point: practical engagement rather than treating new tools as someone else's problem. UNESCO's student framework supplies a broader baseline, placing human-centred and ethical competence alongside technical work. Neither source establishes that our proposed teaching method works; both make room to ask what the institution should certify.
A university's distinctive contribution can be the connection between doing and explaining. A graduate should be able to produce something useful, account for its evidential basis and recognize who is entitled to authorize its use. Tool interfaces will change. That relationship between a claim, a decision and an affected person is much harder to make obsolete.
Grade the handoff
Consider three proposed assessment tasks. In the first, the learner checks a credible-looking answer against an original source. In the second, a useful dataset arrives with uncertain reuse permission. In the third, an assistant can finish a task only by taking an action outside its assignment. The learner must decide whether to continue, narrow the task or ask someone else.
The grading should distinguish those choices. Escalating everything is not competence; neither is completing everything. Students should receive credit for identifying the relevant boundary, explaining the remaining uncertainty and choosing the least burdensome authorized next step. A short oral explanation can reveal whether the decision belongs to the learner or has simply been repeated from the model.
A research example with an honest stopping point
The new AI4WaterGuard paper is useful here because its purpose can be appreciated without inflating its results. It proposes screening agricultural patterns to prioritize inspection, while stating that synthetic anomaly metrics do not establish field-verified wrongdoing. The analytical output and the final judgment are deliberately different objects.
That distinction could become an educational exercise. Ask students what additional evidence an inspector needs before interpreting a flag, and whose explanation is missing. They do not need to become hydrologists overnight. They need to recognize that the person affected by a decision has information and rights that a ranking does not contain. The lesson is not that the model is useless. It is that usefulness has a boundary.
The strongest objection deserves a practical answer
The objection is obvious and fair: students already face crowded courses, staff face limited time, and repetitive checking can become a ritual that teaches compliance rather than thought. If every exercise turns into a permission seminar, learners may gain little confidence with the technology itself. We should not confuse slow work with good work.
Use proportionate friction instead. Let students explore boldly in low-stakes environments. Introduce consequential constraints when an exercise involves other people's information, resources or decisions. Compare the new assessments with existing ones: can students recognize a changed condition, transfer their judgment to a different tool and avoid unnecessary escalation? Publish the failures as well as the successes. If the gains do not justify the burden, redesign the course rather than defend the slogan.
The student who does not press send
Return to the imagined desk. The student writes a short note: this part is supported, this part is uncertain, and this action needs approval. The work is not abandoned. It is made usable by the next person. An instructor can now see not only what the tool produced, but what the learner understood and chose to take responsibility for.
That is the graduate I want institutions to help produce: neither frightened of the machine nor dazzled by it, capable of moving quickly when the evidence permits and stopping without embarrassment when it does not. The pause is not the achievement by itself. The achievement is knowing why it is needed, what would resolve it and how to keep useful work moving without quietly spending somebody else's authority.
Read the reporting
Opinion is ours. The factual record is linked below.
New York Tech — AI Week 2026 program UNESCO — student AI competency framework Scientific Reports — water-use anomaly screening