Why I Stopped Using AI Detection as Evidence – A VP’s Perspective
After the parent-threatening-to-sue incident I described in an earlier thread, our department sat down and formally evaluated our detection-based process. Here’s what we found and why we stopped.
Three findings from our internal review:
1. Legal exposure: any formal academic integrity finding based primarily on an AI detection score would not survive a legal challenge. The tool companies themselves say scores are probabilistic, not forensic evidence. We were using probabilistic data to make formal findings. That’s not a defensible process.
2. Equity pattern: when we audited our flagging history, ESL students, French-language students, and students with writing accommodations were overrepresented. This is a human rights exposure, not just a policy issue.
3. Educational value: the conversation that followed a detection finding – where we asked students to explain their process – was producing the useful information, not the score. The score was just the trigger. We could trigger the same conversation with a simpler rubric for “discuss any section of this essay with me.”
We now use Turnitin for plagiarism only. AI detection is retired from our formal process. Detection scores can prompt a conversation but cannot be referenced in any formal finding.
4 Replies
Join the discussion.
Log In to Reply
The human rights exposure framing is one departments haven't been considering. If your detection process systematically produces different outcomes for students based on language background, accommodation status, or national origin, you have a human rights issue, not just a technology limitation. Every Ontario board should be consulting with their equity officers before deploying these tools formally.
the VP perspective is exactly what's missing from most AI detection debates. when the person with formal accountability for discipline decisions stops trusting the tool, the policy downstream of that tool has to change. the cascade from "detection is unreliable" to "we need different evidence standards" is one more administrators need to be making publicly.
the absence of evidence problem is real: if you stop using detection, integrity incidents that were previously visible become invisible again. the answer isn't to keep using a broken tool - it's to build a system that doesn't depend on it. but that requires redesigning assessment from the ground up, which almost nobody has the time or mandate to do right now.
the equity audit finding is the one that should make every department stop and look at their own flagging history. who are you actually flagging? if the answer is disproportionately your most vulnerable students, that's a policy problem before it's a technology problem.