Design learning experiences students cannot outsource to AI

Students can now ask an AI tool for a competent explanation of almost any concept before we have reached the second lecture slide. Trying to beat that tool on speed or polish is the wrong approach. A university class earns its value by offering something a generated answer cannot: a carefully staged encounter in which students commit to an idea, notice what it fails to explain, revise it with others and reconstruct what they have learned.

This is not entertainment grafted on to serious teaching. It is playfulness with intellectual consequences. A good class can feel like a sequence of small investigations: anticipate, observe, compare, argue, try again. Participation matters in active learning but “activity” alone is not the goal. The task must change what students pay attention to and require them to think before an answer appears.

1. Start with a prediction, not an explanation

Before presenting a finding or theory, give students a case and ask what they expect to happen. Make the commitment small and safe: a 60-second sketch, a choice between two outcomes or a confidence rating from zero to 100. Then ask for the feature that drove the prediction.

In a psychology class, show an ambiguous image before naming the perceptual principle. In a health sciences class, reveal symptoms before the diagnosis. In a law class, pause a judgment before the decision. The point is not to catch students out. An initial attempt prepares them to notice why the subsequent explanation matters; even unsuccessful retrieval attempts can improve subsequent learning when followed by corrective information.

Do not immediately reward the fastest answer. Collect several predictions, including incompatible ones, and let the reveal create a desire for the right answer.

2. Teach through contrast

Students often remember the example but miss the feature that makes it one. Contrast helps perception become analysis. Instead of displaying one flawless model, place two similar ones side by side: a strong argument and a persuasive-sounding weak one: two scans with one diagnostically important difference, two statistical conclusions based on the same result.

• Spotlight guide: Bringing GenAI into the university classroom
• Spotlight guide: Reframe assessment as part of learning
• When we encourage AI use, how can we still assess student thinking?
  

Ask three questions: What changed? What remained constant? Which difference matters? Comparison can prepare learners to understand a subsequent explanation more deeply. It also develops the judgement that AI use now demands. Students need to detect when two fluent responses are not equally valid, not merely produce another fluent response themselves.

Keep the contrast narrow. If everything changes at once, students cannot locate the decisive cue. One carefully chosen difference is often more instructive than five additional examples.

3. Make uncertainty visible

AI-generated prose often sounds certain and final. Expert thinking rarely does. Build moments in which students must expose the boundary of a claim. Ask: “What evidence would change your mind?”, “What else could explain this result?”, “How confident are you now and why did your confidence change?”

Turn this into a low-stakes classroom ritual. Students first record an individual position, then compare it with a partner’s, and finally revise one sentence after hearing a rival account. The revision is the learning artefact. It shows that changing one’s mind is not failure but disciplined responsiveness to evidence.

The teacher should model the same behaviour. Name what the available evidence supports, what it does not establish and where reasonable disagreement begins. Students learn intellectual humility when you show them perception is changeable.

4. End with reconstruction, not recap

A polished final slide can create familiarity but doesn’t guarantee learning. Close the class by removing the explanation. Ask students to reconstruct the mechanism from memory, redraw the model, explain the concept to a sceptical reader or apply it to a case in which details have changed. Retrieval practice is consistently associated with stronger long-term retention than repeated study.

An effective exit prompt has two parts: “What can you now explain?” and “What remains unresolved?” The first requires retrieval; the second preserves curiosity and gives the next class a meaningful starting point. Collecting a few anonymous responses also tells the teacher what actually landed, rather than what felt clear while speaking.

None of these moves requires elaborate technology. Index cards, a hidden slide, paired cases and a blank sheet of paper are enough. AI can still be invited in – after students have made their own reasoning visible – as a source to interrogate, compare or improve.

The question is no longer whether a lecture contains information that students could find elsewhere. Of course it does. The better question is whether the class creates a sequence of mental events that cannot be outsourced: prediction, surprise, judgement, revision and retrieval. When it does, attendance is not about receiving an answer. It is about becoming more capable of judging one.

Emilia Samit is a professor at Nebrija University in Spain.

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.