Bloomberg news reported on several universities in the United States that have started disabling the AI-powered writing detection feature of the Turnitin service, which is typically used to detect plagiarism. However, when Turnitin enabled the AI writing detection feature, it produced high rates of false positives, potentially impacting accused students.
Vanderbilt University criticized Turnitin for releasing this feature without sufficient notice to the university, and claimed that the false positive rate (incorrectly identifying human writing as AI-generated) was only 1%, but failed to disclose the internal workings of the system. When implemented, the system was found to have a significant number of errors, often identifying non-English-native student writing as AI-generated.
Turnitin states that this detection system is merely a tool to assist instructors and does not directly accuse students. Currently, 98% of Turnitin’s client institutions continue to use this feature.
Previously, OpenAI attempted to develop a similar AI-powered writing detection system, but found it to be inaccurate. Even renowned works, such as Shakespeare’s compositions, were mistakenly identified as AI-generated. Even if the system improves to become more precise, students themselves may still be able to adjust their articles to evade detection.
TLDR: Several universities in the US are disabling Turnitin’s AI-powered writing detection feature due to high false positives. Vanderbilt University criticized the lack of notice and the system’s tendency to flag non-English-native student writing as AI-generated. Turnitin claims the system is for instructor support and not intended to accuse students directly. OpenAI’s previous attempt at a similar system also faced inaccuracies, and students may still be able to evade detection in the future.