February 2023
 

A new tool to detect the use of AI-written text and AI chatbots has been released by the developers of ChatGPT
 

ChatGPT, a freely available AI program, has attracted worldwide attention for its ability to create written content based on user prompts that is difficult to distinguish from content written by humans. However, examples of academic content fully created by the tool has shown it has several serious shortcomings, including creating inaccurate or even fabricated citations. Amid concerns that flawed or fabricated research could start to appear in the literature, publishers have been updating their submission policies (see examples here and here).

In response to a backlash against the free use of ChatGPT from many areas of society, its developers have just launched a detection tool to help identify the likelihood of the manuscript being written by Chat GPT. They report that is an “imperfect” tool as yet, but plan to continue improving its reliability.

 
Update (27 July 2023): OpenAI, the developers of ChatGPT, have now removed their AI detection tool, citing its “low rate of accuracy”. This was announced as an update to their original blog post about the tool. The development of reliable AI- detection tools remains a challenge: for example, a recent study in the journal Patterns found a false-positive rate of 61.3% in widely used detection tools when assessing papers written by non-native English authors.