The ChatGPT language system, which works with artificial intelligence, has caused a sensation in recent weeks, but also raised concerns.
For example, that schoolchildren and students use the software to cheat.
Open AI, the maker of ChatGPT, has now released a technology that is supposed to recognize whether a text was written using the software or other AI systems.
Such tools could help when AI is used for "academic dishonesty" or propaganda, the company said.
Business correspondent in New York.
Follow I follow
Like ChatGPT itself, the detection tool is freely available.
However, Open AI admits that it is "not completely reliable".
It was only 26 percent correctly identifying AI as the likely source of text, and it was 9 percent identifying incorrect human-written text as AI.
So far, it has only been recommended for use with English-language texts; it performs worse in other languages.
It is also unreliable in detecting programming code written with AI.
In addition, it can be easily circumvented by slightly modifying AI texts.
It is also not able to detect plagiarism.
Basically, the longer the texts are, the more reliable it is.
Open AI says the technology shouldn't be used as a "primary decision-making tool," but as a complement to other methods of getting to the bottom of the source of text.
Open AI wanted to make the system publicly available to see if "faulty tools" like these are also found useful.
The technology only works with texts that have at least 1000 characters.
It classifies the entered material into five categories: As likely, possibly, unlikely or very unlikely written with AI or as unclear.
The system was trained with pairs of texts on the same topic, one written by humans and the other by AI.
Mainly texts written by adults were used, which is why the technology is often wrong with texts by children.
In its communication, Open AI points out that the recognition of AI texts is particularly relevant in educational institutions.
In fact, since the release of ChatGPT at the end of November, passionate discussions have been going on in schools and universities on how best to deal with such technologies.
In some cities, such as New York, access to the software on public school computers has been blocked.
Many teachers and professors are also considering ways to integrate ChatGPT into the classroom.
In addition to Open AI, there are other companies developing technologies to recognize AI texts.