OpenAI recently released a promotional blog post offering insights for educators. The post highlights how some teachers are incorporating ChatGPT as an educational tool and provides suggested prompts for initiating its use.
In a related FAQ section, OpenAI officially acknowledges what has been widely recognized: AI writing detectors are ineffective, despite their frequent use in penalizing students with false findings.
In response to the question “Do AI detectors work?”, OpenAI said:
In short, no. While some (including OpenAI) have released tools that purport to detect AI-generated content, none of these have proven to reliably distinguish between AI-generated and human-generated content.
Back in July, ArsTechnica showed how AI tools including the popular GPTZero don’t work with experts even calling them “mostly snake oil”.
AI tools often produce incorrect results because they rely on unproven detection metrics. In the end, there is no foolproof way to consistently differentiate AI-generated text from human-written text, as detectors can be circumvented through paraphrasing.
In that same month, OpenAI discontinued its AI Classifier, an experimental tool designed for identifying AI-generated text, due to its dismal 26% accuracy rate.
Even ChatGPT Can’t Tell
The FAQ even goes on to say that ChatGPT itself is not capable of detecting any AI-written text. Here is what the company said:
Additionally, ChatGPT has no ‘knowledge’ of what content could be AI-generated. It will sometimes make up responses to questions like ‘did you write this [essay]?’ or ‘could this have been written by AI?’ These responses are random and have no basis in fact.