OpenAI has recently launched a new AI text classifier aimed at detecting AI-generated text. The classifier is designed to distinguish between text written by a human and text written by AI from various providers, including ChatGPT and other OpenAI models.
The classifier is aimed at addressing the issue of false claims made about AI-generated text being written by a human. OpenAI hopes that it can be used to help mitigate automated misinformation campaigns, academic dishonesty using AI tools, and cases where AI chatbots are positioned as humans.
Update: OpenAI has unfortunately discontinued the AI text classifier tool. Due to a “poor rate of accuracy”, the team has ultimately decided to shut down the tool and refocus their efforts elsewhere. You can read more about this change on the Originality.ai blog post.
Reliability of the Classifier
The detector has been trained on a dataset of human-written and AI-written text on the same topic. While the classifier is not perfect, it has been shown to have improved reliability compared to previous versions. In tests on a “challenge set” of English texts, the classifier was able to correctly identify 26% of AI-written text as “likely AI-written” while incorrectly labeling human-written text as AI-written 9% of the time. It’s worth noting that the reliability of the classifier increases with the length of the input text.
The OpenAI text classifier has a number of limitations, and it is not meant to be used as a primary decision-making tool. It is unreliable on short texts (below 1,000 characters), can sometimes confidently label human-written text as AI-written, and performs poorly in languages other than English. Additionally, the classifier struggles with text that is very predictable, and AI-written text can be edited to evade the classifier.
Impact on Educators
OpenAI recognizes that identifying AI-written text is an important issue for educators. The company is reaching out to educators in the US to learn about their experiences and to gather feedback. OpenAI has created a preliminary resource on the use of their language model, ChatGPT, for educators, which outlines the capabilities and limitations of AI generated text classifiers in the classroom. They are inviting those directly impacted by these issues, such as teachers, administrators, parents, students, and education service providers, to provide feedback using a form on their website.
The launch of this new AI classifier is an important step in addressing the challenges posed by AI-generated text. While the classifier has limitations, it is a work-in-progress, and OpenAI will continue to develop and improve it. By engaging with educators and those directly impacted by these issues, the company hopes to create a safe and responsible deployment of their large language models.
You can check out the work-in-progress build of the tool here:
https://platform.openai.com/ai-text-classifier (no longer available).