OpenAI has released a tool called AI Text Classifier which aims to detect whether the input text was generated using an AI tool such as ChatGPT.
AI Text Classifier is a fine-tuned GPT model that predicts the likelihood of AI-generated text. However, the accuracy of the tool is limited and not completely reliable. According to the company’s evaluation, it correctly identified 26% of the texts written by AI as “highly likely to be written by AI”, and 9% of the time, it mislabels human-written text as AI-written. The tool’s reliability increases with longer text inputs and is significantly more reliable on the text generated by current AI systems.
While the tool provides a quick way to determine if AI generated a text, there are limitations to its use. It requires a minimum of 1000 characters, and the accuracy can vary, especially for text written by children or in languages other than English. Additionally, AI-generated text can be easily edited to evade detection.
OpenAI’s focus with this tool is education. As AI-generated text becomes more prevalent, identifying text written by AI has become a point of contention among educators. OpenAI is engaging with educators in the US to understand how AI-generated text, such as ChatGPT, can be used in the classroom and is soliciting feedback from educators.
OpenAI is not the first company to develop a tool for detecting AI-generated text. However, it aims to further the conversation on how AI-generated text should be used and regulated in education. The tool can be accessed with a free OpenAI account and is as simple as pasting the text into a box and clicking a button to see the results.
