OpenAI, the artificial intelligence (AI) company behind the ChatGPT chatbot, has quietly pulled its tool specifically designed to detect AI-written text due to its “low rate of accuracy” just months after it was made public.

The tool — known as a “classifier” — was launched in January this year to help with distinguishing between human-written text and text written by generative AI.

OpenAI explicitly stated in the tool’s original launch post that the classifier was “not fully reliable,” and that, in evaluations, the tool correctly identified 26% of AI-written text as “likely-AI written,” while incorrectly labelling human-written text as AI-written text 9% of the time.

Due to many issues — the tool was “very unreliable” on shorter (1,000 characters and fewer) texts, and it was “sometimes extremely confident in a wrong prediction” — the company warned its users to not use the classifier as a primary decision-making tool.

In a recent update, the firm wrote that the AI classifier is “no longer available due to its low rate of accuracy,” and that they’re “working to incorporate feedback and are currently researching more effective provenance techniques for text.”

OpenAI’s pulling of their own tool further highlights the difficulties in classifying what’s been created by generative artificial intelligence and what hasn’t amid the rapid surge in its usage.

In the field of education, the introduction of ChatGPT and similar tools has sparked fears over how it could be used by students, academics, and the like to cheat and plagiarise.

Recommended

• Major AI Players Form a New Forum for ‘Responsible AI’
• AI: Cyber-friend or Cyber-foe?
• Edinburgh Named the UK’s most AI-ready City – Outside London

In a move to call attention to the believability of text written by generative AI, a group of academics submitted a paper earlier this year titled “Chatting and Cheating: Ensuring Academic Integrity in the Era of ChatGPT,” according to The Guardian. After peer review, the paper was published in an education journal; it was later revealed by the academics that the true “author” was ChatGPT.

As schools, universities, and other educational institutions across the globe turn to AI detection tools to thwart attempts at cheating and plagiarism, there are growing concerns over their reliability.

Last month, Rolling Stone profiled a University of California, Davis, student who was wrongfully accused of using generative artificial intelligence for a written assignment. The university had “early access” to the detection tool in question — the developer had noted it had “a small risk of false positives.”

“I think they just need to be more careful with how they approach it,” suggested the student.