Hot take: AI art detectors are doing more harm than good in classrooms

I was proctoring a digital design midterm at my kid's school last month when the plagiarism checker flagged a student's original 3D model as AI generated. The kid spent 12 hours on that mesh in Blender and the algorithm couldn't tell the difference between a human-made torus knot and a diffusion model output. The teacher was ready to give him a zero until I pulled up his undo history in Blender. That moment made me realize we're punishing actual effort because these detection tools have like a 70% false positive rate on complex work. Why are we trusting software that can't even distinguish between a student's 40 hour project and a 30 second prompt? Has anyone else seen this backfire in their field?

3 comments

3 Comments

pat_moore1mo ago

Why not just have students record their screen the whole time?

cooper.drew1mo ago

Man that's rough, @pat_moore but screen recording doesn't always catch things either - I've seen students who genuinely code or model for hours get flagged because the detector picks up on clean mesh topology or consistent brush strokes. Real talk, the better fix is having a quick process check where the student walks through their project history or version log with the teacher. Takes five minutes and beats trusting a black box algorithm that punishes the kids who actually put in the work.

michaelgrant1mo agoTop Commenter

@pat_moore I just ask them to show me layer history in their file. Works every time.

-1