AI Detectors: CTL's Recommendations and Cautions

Despite their promise, applications designed to detect writing produced by generative AI are unreliable. The two readings in this section provide different perspectives on this topic: our RWS colleague Jacob Hubbard writes an informal (at times playful) essay with a serious conclusion; and four researchers at Stanford (in computer science, data science, and electrical engineering) share the results of research into the bias demonstrated by AI detectors and conclude with some recommendations.

Here at SDSU, the AI content detection feature of Turnitin.com will remain off for the time being. Many CSU campuses disabled this feature last Spring, including our own; so did many other colleges and universities across the country. The SDSU Senate's Instructional and IT Committee recommended that ITS keep the feature turned off. The reason is that Turnitin’s AI detector is prone to errors, in common with similar applications.

Faculty turning to AI detectors that they find on the internet are advised to use caution. False positives and false negatives are not unusual. Though CTL does not recommend AI detectors, if you still intend to use an AI detector, CTL recommends mentioning that in your syllabus (along with any other required technology). And we still do not know the full ramifications for student privacy in the use of this technology. Consider using the results of an AI detector as the starting point for a discussion with a student.