AI in Education: College Assessment's Steroid Era

Cheating with AI has become the norm

AI may not have dramatically changed most people’s lives yet, but it has disrupted college learning and assessment by turbo-charging cheating. One Brown economics class averaged 96 on a take-home midterm and 48 on an in-person final — the same students who submitted perfection floundered when tested on the same material without access to AI.

The rate of cheating is a property of the assessment structure rather than an indictment of student character, which is why academic integrity can be rebuilt — but only if educators make consistent changes in how they prepare and assess students. Honor codes are not enough in the new AI Era: any assessment with significant grade weight requires supervision. Formative practice must be encouraged by rewarding completion rather than accuracy, so it is worth doing honestly. And graded assessments should draw from the same material as the practice, so the honest path and the high-scoring path become the same path. When students know their practice aligns directly with a test AI cannot help them pass, the reason to shortcut the practice mostly disappears.

To support this structure, PracticeQuest generates question variants from expert-written templates, creating a limitless source of non-repetitive practice that can also supply questions for proctored tests. In the general biology course redesign this approach came from, the DFW rate fell from about 35 percent to about 22 percent across my own sections and those of colleagues. download the paper for a full analysis of the problem of AI in Education

PracticeQuest AI in Assessment whitepaper.pdf

Download PDF • 197KB