LLumen
45 minutes

Designing AI-Aware Assignments

Rewrite tasks so AI either cannot fake the learning or is explicitly authorized with verification requirements.

Objectives

  • Classify assignments as AI-resistant, AI-authorized, or AI-risky
  • Add process evidence requirements without busywork
  • Pilot one redesigned task next week

Three redesign moves

1) Local/personal data the model cannot invent credibly. 2) Process checkpoints (outline conference, oral, lab notebook). 3) Explicit AI roles (critique only, never draft thesis) with disclosure.

What not to do

Do not rely on AI detectors alone. Do not assign “write 5 pages on X” unchanged. Do not ban tools while modeling secret use yourself.

Try this week

Take one upcoming assignment. Label it resistant / authorized / risky. Rewrite with one process artifact and one verification requirement. Share in PLC.