San Francisco-based Open AI launched ChatGPT in November 2022. The chatbot uses a generative AI model to produce text in response to questions.  

OpenAI’s new classifier has been made publicly available, but has “a number of important limitations”, the company said.

In a blog post authored by staff members, OpenAI said: “While it is impossible to reliably detect all AI-written text, we believe good classifiers can inform mitigations for false claims that AI-generated text was written by a human: for example, running automated misinformation campaigns, using AI tools for academic dishonesty, and positioning an AI chatbot as a human.”

In evaluations carried out by the company using a set of English texts, the tool correctly identified 26% of text written by AI as ‘likely AI-written’, but incorrectly labelled human-written text as written by AI 9% of the time, OpenAI said.

The tool is an update on a classifier previously launched by the firm, which said its reliability “typically improves as the length of the input text increases”.

However, the company stated the new tool should not be used as a primary method of determining the source of a piece of text, but rather as a complement to other methods.

Concerns have been raised about the potential for ChatGPT and other generative AI models to be used to plagiarise content, for example, for student essays.

Researchers have discussed the potential for the tool to influence their work, including analysing qualitative data at scale, but also shared concerns that it may proliferate false information. 

OpenAI is also launching a paid pilot subscription plan for ChatGPT in the US, that will give subscribers access to the tool during busier periods and offer faster response times.  

The company said that the tool will continue to be free to access.