If We Can’t Act on Detection Results, Why Are We Using Them?
here’s the contradiction I’m sitting with.
Admin says: use AI detection tools to flag potentially AI-generated student work.
Also admin says: you cannot formally accuse a student of academic dishonesty based solely on a detection score.
So what exactly are we doing with these tools? we’re generating evidence we can’t use, having conversations we can’t document, and flagging students based on scores that we acknowledge aren’t reliable enough to act on. And in the meantime, students who didn’t use AI are sometimes getting flagged and are understandably frustrated.
i get that the technology is imperfect and schools are trying to find a middle ground. but the current middle ground is “collect data, don’t use it, pretend you’re doing something.” is anyone else operating this way? and has anyone found a policy framework that actually resolves the contradiction?
4 Replies
Join the discussion.
Log In to Reply
We resolved it by being explicit: detection scores are not evidence, they're a trigger for a conversation. The conversation itself - asking the student to walk through their process, explain their sources, discuss their choices - is where you actually figure out what happened. Turnitin gives you a number that says "worth investigating." Nothing more.
Ray's framing is exactly the one our board eventually landed on after a lot of debate. Detection as a trigger for conversation, not as evidence for a finding. The academic integrity case has to be built on the conversation: does the student understand their own work? Can they explain their choices? Are there inconsistencies between their process and their product?
The boards that are getting this right have moved away from detection-first to process documentation. They're asking students to submit drafts, annotations, research notes alongside final products. The work itself demonstrates understanding in a way a detection score never can.
My honest view: AI detection will never be reliable enough for high-stakes academic integrity decisions. the technology is in an arms race with humanization tools. GPTZero catches something, Originality.ai catches something different, and humanizers evolve to beat both. the policy should not be built around a technology that is designed to be circumvented.
from the admin side: agreed. we've told teachers the score isn't actionable without additional evidence, and now half of them are asking why we bought the subscription. valid question, honestly.