AI text detection tools are really easy to fool

Daphne Ippolito, a senior research scientist at Google specializing in natural language generation who also doesn’t work on the project, raises another concern.

“If automated detection systems are to be used in an educational setting, it is important to understand their false-positive rates, as falsely accusing a student of cheating can have serious consequences. for their academic career,” she said. “The false-negative rate is also important, because if there’s too much AI-generated text that looks like human handwriting, the detection system won’t be useful.”

Compilatio, the company that makes one of the tools tested by the researchers, says it’s important to remember that its system only points to suspicious passages, classified as potential plagiarism or content. potentially AI-generated.

“It is up to schools and teachers to mark analyzed documents to validate or attribute the knowledge that the author of the material actually obtained, for example by introducing means of investigation. additional—questioning orally, additional questions in a controlled classroom setting, etc,” said a Compilatio spokesperson.

“In this way, the Compilation tools are part of a teaching approach that really encourages learning about good research, writing, and citation methods. The spokesperson added that the compilation software is an editing tool, not an editing tool. Turnitin and GPT Zero did not immediately respond to requests for comment.

We’ve known for some time that AI-written text detection tools don’t always work the way they should. Earlier this year, OpenAI announced a tool designed to detect text generated by ChatGPT, admitting that it only marks 26% of AI-written text as “likely AI-written”. OpenAI has directed the MIT Technology Review to a section above website for educators to consider, which warns that tools designed to detect AI-generated content are “not foolproof.”

However, such failures don’t stop companies from launching products that promise to do the job, said Tom Goldstein, an assistant professor at the University of Maryland who was not involved in the study.

“Many of them aren’t very accurate, but they’re not quite a disaster either,” he added, noting that Turnitin managed to achieve some detection accuracy at the rate of false positives are quite low. And while studies that shed light on the shortcomings of so-called AI text detection systems are important, it would be helpful to expand the scope of research to AI tools beyond ChatGPT, Sasha Luccioni , said a researcher at AI startup Hugging Face. .

For Kovanović, the whole idea of ​​trying to detect AI-written text is misguided.

“Don’t try to detect AI – make AI usage not the problem,” he said.

Source by [author_name]


News7h: Update the world's latest breaking news online of the day, breaking news, politics, society today, international mainstream news .Updated news 24/7: Entertainment, the World everyday world. Hot news, images, video clips that are updated quickly and reliably

Related Articles

Back to top button
Immediate Peak