3 Comments
User's avatar
David Martino's avatar

Cross-examination is one test no tool can take for the witness. A witness who memorized a prepared account often breaks on the second follow-up question. A student who memorized a prepared essay faces the same risk in a seminar discussion.

Adelina Raheel's avatar

I'm a student and a writer, and I use AI, so I read this half nodding and half defending myself. Where you're right, you're very right. The struggle is the learning. Nobody gets stronger watching someone else lift. The wedding speech question is the real cost to me, because that doubt is already leaking onto things people did write themselves. So maybe the answer isn't only keeping the tool out of the room. What if we graded the process again, the drafts and the crossings-out, and not just the finished page?

Ryan McCarl's avatar

"Nobody gets stronger watching someone else lift." ← Great analogy for why using AI for schoolwork is self-defeating.

Re: alternative grading — all grading is problematic. (Abolishing grades is also problematic in different ways.) But I am especially skeptical of the "grade the process" approach because it's so easy to game such grading schemes. "Grading the process" has been a common approach to grading for many decades, long predating AI. Students can fake the process even more easily than they can fake the outcome, once they learn what criteria instructors are looking for. Grading the process also introduces more subjectivity into grading, encourages micromanagement of students' workflows based on instructor assumptions that may not reflect how subject-matter experts actually work, and unfairly penalizes students who produce better outcomes in unorthodox but legitimate ways.

I'm not sure what the solution is here, but I would start with teaching students that using AI on homework entirely defeats the purpose of school along with any joy and benefit they might gain from learning; flatly banning its use in classrooms and for schoolwork, with exceptions as needed for specialized computing courses and AI-related electives; and enforcing the ban with Pangram and, eventually, watermarking. Train a panel of administrators and students to distinguish AI vs. human-written text; the professor gets a vote, the panel gets a vote, and Pangram gets a vote, and if both the humans and the AI detector think you cheated, you redo the assignment or sit for an oral exam questioning you about the work. If it keeps happening, the penalties escalate. That's one idea, anyway. One thing I'm sure of is that schools should disincentivize its use rather than encourage it.