Will new software be able to detect whether something was written by a bot or a human?

Reserach about the ChatGPT

I have been reading a lot about ChatGPT lately and decided that the question to ask is; how do students use this in a good way? Is it necessarily all bad if students use this? Can they generate ideas, ask for help with vocabulary, and make summaries of a topic? Those are good ways to use this. But there is a dark side here.  In other words, how to detect if students are claiming that the writing by ChatGPT is their own?

In the discussions about ChatGPT, you will find a lot of articles about how students have been cheating since the early days of academia. Yes, we all know that. We had a lot of conversations here at our school when Google translate became the students’ best friend in French, German, or Spanish classes. Somehow the teachers would always know. And, of course, now, you can translate your text while writing in Microsoft Word. And who knows how long it will take before Microsoft Word has the AI function inside their products. Every student will most likely be equipped with a personal writing assistant.

Anyway I have been reading about Turnitin’s new model, and I quote here (The Hechinger report) :

Turnitin’s model compares submitted writing to other writing available on the internet, including archived student papers, academic journals and other sources. ChatGPT and generative AI pose a new challenge, because there is nothing to compare the submitted samples to, she said.

Turnitin is working on new software that will be able to detect whether something was written by a bot or a human. The software will be able to tell because bots write differently than humans do. Instead of writing based on context, as you or I would, a bot writes word by word, predicting what should come next based on what has already been written. A bot would pick the most statistically average word, whereas a human might pick a word with more flair, she said.

The new software is likely to be available within the next six months, Chechitelli said.

“We’re entering a new chapter.”

Annie Chechitelli, chief product officer at Turnitin, a plagiarism detection service used by many colleges and universities. 

Besart Kunushevci was thinking about the possibility of something like ChatGPT long before it launched. He’s the founder and CEO of Crossplag, a company that began as a multilingual plagiarism checker, similar to Turnitin. He said his software can tell if, for example, a student copied something from an Italian research journal, translated it into English and pasted it into an essay they turned in. The program is similar to Turnitin’s.

Kunushevci said his software can detect whether submitted text was written by a human or a bot, co-written by both, or even if the student used a program to paraphrase the bot-written text (presumably, an attempt to throw off this type of checker). I tested it by copying a draft of this article into the checker. The system gave it a score of 1 percent and said it was likely written by a human. When I copied in a short story about squirrels from ChatGPT, it received a score of 94 percent, meaning “This text is mainly written by an AI.”

Next, Kunushevci said, he’s hoping to create a system that can learn the style of one person’s writing, and then detect if something is submitted that doesn’t seem to be written by that person. With Crossplag, he said he is trying to eliminate academic dishonesty, one pesky form at a time, even as it evolves.

I would love to hear from you