OpenAI has quietly discontinued its AI Classifier, dashing the hopes of educators who were relying on it to detect content created using generative AI tools. The decision was made due to the tool’s poor accuracy, which led OpenAI to unplug the AI detection tool last week without making a formal announcement. The company acknowledged the limitations of the classifier, including its unreliability with texts under 1,000 characters and mislabeling human-written text as AI-generated. Despite the initial claims of distinguishing between human and AI-written text, the classifier’s performance fell short, correctly identifying only 26% of AI-generated text while mistakenly labeling human-written text as AI-written 9% of the time.
The education sector, in particular, was keenly interested in accurately detecting AI-generated content, especially after the launch of ChatGPT, which raised concerns about students using the AI chatbot to produce essays. OpenAI expressed awareness of these concerns and stated a commitment to exploring more effective techniques for text provenance. However, the AI detection tool’s discontinuation signals that the current approach did not yield satisfactory results. OpenAI continues to learn from this experience and intends to enhance its efforts to develop and deploy mechanisms that enable users to identify AI-generated audio or visual content. Despite the setback, new tools for detecting increasingly sophisticated AI-generated content continue to emerge, creating a growing field of AI detectors. As of now, OpenAI has not provided a response to media requests for comments on this development.