OpenAI’s ChatGPT is too good for its own AI to detect

OpenAI has shut down Classifier, its AI tool to detect AI-written work, because of what we discovered at its launch: It stinks.

When OpenAI launched Classifier in January, it was supposed to have been smart enough to detect when a user (say, a student) used ChatGPT or another AI program to author a work. OpenAI pitched Classifier as a way for a teacher to identify if an essay, for example, had been authored by AI instead of as an original work.

But the tool struggled when we tested it, including identifying famous works as AI-authored. Instead of fixing the problems, OpenAI has apparently just shut it down. The Classifier page is blank, and OpenAI added this to the original announcement of the Classifier tool:

“As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy. We are working to incorporate feedback and are currently researching more effective provenance techniques for text, and have made a commitment to develop and deploy mechanisms that enable users to understand if audio or visual content is AI-generated.“

At the time Classifier launched, OpenAI said that the classifier identified 26 percent of AI-authored text as authentically human, and deemed 9 percent of text written by a human as AI-authored.

Our tests previously found that a rival service, Hive Moderation, seemed to work. The tool is a free AI detection tool which performs the same function, but returns a confidence percentage, or the likelihood that the text sample contains AI text.

Students have loudly complained online that even when they do write original essays, flawed AI detection apps can trip them up. With honor codes at universities and high schools allowing little latitude for plagiarism, one false positive can doom a student’s academic career. It’s possible that Classifier’s demise will be used as a reason to reconsider how educational institutions deal with the issue.