Assignment Redesign: From AI-Proofing to AI-Resilience

AI-proofing an assignment enters an arms race it cannot win: capabilities drift, false confidence sets in, enforcement lands unevenly, and task design turns brittle. This lecture replaces that goal with AI-resilience — an assignment whose evidence of learning stays meaningful whether or not students use AI — built through three strategies: process visibility, use clarity, and authentic context. The redesign triangle keeps the learning goal, the AIAS level, and the evidence form pointing at the same outcome, with the lock-versus-blueprint contrast as the mindset shift.

How to make student thinking visible without drowning anyone in paperwork. The one-sentence test — what learning decision does this element reveal? — separates evidence from compliance, and a workload audit (about thirty minutes of student time and ten minutes of review per submission as warning thresholds) keeps the element honest at real enrollments. The rule of one says add at most one process-evidence element per assignment, and the lecture warns against the surveillance trap of demanding complete prompt logs, offering selected excerpts plus a brief reflection instead.

When AI is integrated on purpose, vague permission defaults students into Oracle use. This lecture specifies three designed postures — Tutor (questions before answers, hints after documented attempts), Adversary (flawed output the student critiques; the critique is the submission), and Collaborator (student drafts first, AI suggests, student documents accept/reject decisions) — each matched to AIAS levels and protected by a boundary with a trigger, a documentation response, and a restorative follow-up. A free-tier check confirms the assignment works without paid access before it ships.