A HARVEST KERNEL WORKING SESSION
The AI Judgment Assignment
Take one spring course and build one assignment where students judge AI output by your field’s own standard. You name the task AI already does for your students, the standard your discipline checks first, the outcome, the assignment steps and a three level rubric row. You leave holding a finished assignment you can paste into your course. You do the thinking. Nothing here writes it for you, and nothing you type leaves this page.
Why this exists. AI majors nearly doubled in a year, and they are training people to build AI. Every one of them is a program a student has to choose. Everyone else meets AI drafts in their own field, and the person qualified to teach them when a draft is reliable is the person who already grades that kind of work. That is you. One outcome inside the course you already teach reaches every student in it.
SEED
Pick the course and the task AI already does
One course. Then the task your students already hand to AI, not the one you wish they would not. A summary, a first draft, a worked solution, a lab discussion, a citation list. Be honest about it. The assignment only works if it meets students where they already are.
SPROUT
Name the standard your field checks first
When an expert in your discipline looks at that kind of work, what do they check before anything else? The source. The calculation. The protocol. The evidence for the claim. Write it in the words you use when you grade.
GROW
Write the outcome, then the assignment
The outcome is one sentence. Press the button to start it from your task and standard, then make it sound like you. The assignment follows from it: students get an AI draft of the task, mark every place it fails the standard, show the evidence for each mark, and hand in the corrected version beside the marked one.
HARVEST
Write one rubric row, three levels
At the top, the student found the failures that matter and proved each one. In the middle, the student found some and asserted more than they proved. At the bottom, the student polished the AI draft without testing it. Press the button to start the row from your standard, then rewrite each level in your grading voice.
0 of 11 fields filled
