How to Build Your Own AI Detector

Student submissions to college-level courses have shown increasing amounts of artificial intelligence (AI) content over the last few years. This happens in spite of the traditional academic integrity values that student submissions should be the product of their own efforts. Also, AI use is often not disclosed, raising another authorship problem. There is tremendous cultural pressure to do everything faster and with less effort, even if it undermines learning processes.

Using AI detection technology has been controversial. It works, but it works imperfectly. This imperfection raises concerns about false positive errors: Mistaken judgments of AI use when, in fact, the students have written their own work. My view is intermediate. AI detection has a place in teaching when it is used with good judgment. More details are in my 2024 paper that provide support for this cautious use position.

There is also the practical problem of which AI detection technology to use. Free trials are limited and a headache to use. My institution has made very little effort to support faculty through providing technology. I don't have many tools available. All I've got is Microsoft Copilot (through our Office license) and Turnitin.com's service that is limited to assignments.

My solution is to make my own AI detector. I'm a DIY kind of guy, so this was a natural to me.

The following link has a prompt that is given to general AI. This prompt can turn generative AI into an AI detector. This allows me to run AI detection when needed for special situations. I'm sharing it with colleagues in case it might be helpful in some way.

How to Build Your Own AI Detector