Do something impossible with AI

Recommended for: Advanced students, in a course where persuasion, communication or design is assessed

Background for the instructor

Most AI assignments ask students to use a tool on a task they could already do. This one does the opposite: it starts from something they believe is beyond them, which is the only condition under which they find out where the tool actually stops. Students discover the limits quickly, and the skill they practise is pivoting — trying something, watching it fail, and adapting.

The assignment does not work cold. It needs smaller scaffolded tasks earlier in the course so students already know the range of things AI can be asked to do.

Preparation

Run two or three short AI-involving exercises in the weeks before, so students have a repertoire rather than a blank prompt box. Be explicit that the grade does not depend on the impossible task succeeding.

Procedure

1. Each student names a persuasive task they would not trust themselves to accomplish — a case they cannot make, an audience they cannot reach, a register they cannot write in.
2. Before starting, the student writes down how they expect to use AI: for what, at which point, and why they think it will help.
3. They carry out the task, keeping a record of what they actually did with the tool as they go, including the attempts they abandoned.
4. They present the result, and report where AI helped, where it did not, and where they had to change course.
5. Close by comparing across the class: which kinds of impossible turned out to be tractable, and which did not.

Discussion questions

• Where did your actual use of AI diverge from the plan you wrote at the start?
• What could the tool not do that you had expected it to — and what could it do that surprised you?
• At what point did you stop asking the model and start thinking it through yourself?

Assessment opportunity

Assess the gap between plan and practice. A student whose task failed but who can say precisely why has met the learning outcome; one whose task succeeded without their being able to account for it has not.

Reference

• Kiera Allison, Assistant Professor of Commerce, McIntire School of Commerce, University of Virginia, and 2024–25 Faculty AI Guide. "Do Something Impossible with AI", UVA Teaching Hub. teaching.virginia.edu
• Discussed in The Norton Guide to AI-Aware Teaching (Vee, Watkins & Bruff, W. W. Norton, 2026).