OpenAI, creator of Chat GPT, has clearly heard criticisms of its AI chatbots. Namely, from an educator who expressed concern about students using her ChatGPT to do homework or write reports.
Tuesday, OpenAI release(opens in new window) A new AI classifier tool aimed at directly addressing these concerns. This free web tool was created by OpenAI to help users determine if a block of text was written by a human or a computer.
Using the tool is easy. Just access the AI classifier(opens in new window)Click to log in, then copy and paste the block of text you want to see. The classifier then ranks the texts on a scale of Likely, Probable, Unclear, Unlikely, or Very Unlikely generated by AI.
Copy and paste text to find out if it was written by a human.
Credit: OpenAI / Screengrab
While this seems like it would be very helpful in addressing some concerns about AI-written text, it’s very clear that OpenAI has caveats. For example, according to OpenAI, “classifiers are not completely reliable.”
“In an evaluation against a ‘challenge set’ of English texts, the classifier correctly identified 26% of texts written by AI (true positives) as 9% incorrectly labeled text as written by AI.(False positives) at the time,” the company said. blog post(opens in new window).
3 Things to Know Before Talking to ChatGPT About Mental Health
There are other limitations to the classifiers addressed in the OpenAI announcement. For example, the classifier does very poorly on texts with less than 1,000 words. Human-written content can be mistakenly identified as computer-generated text. And it only works well with English text.
OpenAI isn’t the first to release tools to detect when text is computer-generated. In fact, it’s not the first classifier released by the company itself. However, OpenAI says it is “significantly more reliable” than previous tools. And it is quite noteworthy that such tools are continuously being developed by one of his leaders in the AI field.
Still, it will take some time before teachers and educators can confidently rely on AI detection tools to investigate misbehavior in the classroom.
