OpenAI has recently released a classifier that aims to determine whether a piece of text was authored by artificial intelligence (AI) platforms like its own ChatGPT tool.
The company released the tool after various educational institutions and school districts banned ChatGPT because some students are relying on it completely to write their papers and pass it off as their own, which of course is cheating. Currently, ChatGPT is banned in New York City, Seattle, Los Angeles, and Baltimore public school districts. Some universities in France and India have also restricted access to the tool. Finally, some states in Australia have blocked students from accessing ChatGPT on school internet networks.
OpenAI describes its text classifier as "a fine-tuned GPT model that predicts how likely it is that a piece of text was generated by AI from a variety of sources, such as ChatGPT." Despite this claim, however, the company itself admits that the tool is unreliable. In their own evaluation of English texts, the classifier correctly identified only 26% of AI-written text as likely written by AI, and deemed 9% of human-written text as AI-authored. What's more, OpenAI says that the classifier might be unreliable on texts that are below 1,000 characters and written in other languages aside from English.
In our own testing, OpenAI's classifier correctly deemed most of the articles published on Neowin as "very unlikely" to be generated by AI. However, the tool was indecisive with our recent Nothing Phone (2) coverage, saying that it is "unclear if it is AI-generated." When tested with content generated via ChatGPT, the classifier seems to be a bit doubtful, saying that the content is "possibly AI-generated."
This is probably why OpenAI says that the results generated by the classifier should not be the "sole piece of evidence" when determining if a piece of content was written by AI. Thankfully, there are other tools you can use. For instance, Stanford researchers recently introduced DetectGPT, a tool to help educators detect AI-generated papers. Also, a computer science student at Princeton developed a similar tool that can "quickly and efficiently" determine whether an essay is a created by ChatGPT.
