Paris - (AFP) - The artificial intelligence program "ChatGBT", which is able to write texts in response to simple questions, has spread widely in the world of education, prompting teachers to question the appropriateness of banning it or benefiting from it.

In mid-December, just weeks after the tool was made available by California startup OpenAI, eight Australian universities announced that they would adjust their exams, classifying students' use of AI as cheating.

In this context, students' exams will be "monitored" in 2023 by "increasing the use of paper and pen," said Vicky Thompson, president of the "Group of Eight (universities)" via a blog affiliated with The Australian newspaper.

Recently, after several media outlets reported on increased use of the tool by students around the world, encouraged especially by videos on TikTok, public schools in New York restricted access to ChatGBT on their networks.

Gina Lyle, a spokeswoman for the city's education department, told AFP that the tool "doesn't help develop critical thinking and problem-solving skills, which are skills necessary for success academically and in life."

ChatGBT is a chatbot that has been "trained" by fumbling huge amounts of data gleaned from the Internet, and can "predict" the likely sequel to a text. However, due to the absence of human reasoning, this bot produces a remarkable mixture of correct answers and factual or logical errors that are somewhat difficult to detect.

It happens, for example, that the robot includes the whale shark among marine mammals, or mistakes the area of the countries of Central America, or “forgets” some historical events such as the Battle of Amiens in France in 1870, or fabricates references that do not exist in the original.

However, in the world of education, there are some voices calling for this innovation to be integrated into the teaching media.

Antonio Caselli, a professor at the Polytechnic Institute in Paris and author of a book entitled "Waiting for the Robots" ("Waiting for the Robots"), told AFP that "ChatGPT" is an innovation "important, but not more important than calculators or text-editing tools" that have found a place for them. Finally at school.

"ChatGPT+ can help with a first draft when you find yourself facing a blank piece of paper, but after that, you have to write and style" the scripts, he added.

The expert also noted that ChatGBT partly reflects the teaching philosophy, based on the teacher asking the questions.

This time, it is the student who is questioning the machine, "and it is an opportunity for us to see how the students are carrying out the tasks assigned to them, to make them work on fact-finding, and to check whether the bibliographic references generated are correct," says Caselli.

According to Olivier Ertscheid, a researcher in information sciences at the University of Nantes (western France), banning the tool is "counterproductive" because it enhances students' desire to benefit from it.

As with the arrival of Wikipedia or search engines, the challenge for teachers, he says, is to "test the limits" of these tools.

Finally, the response required to detect texts generated by AI techniques is regulated. For example, the online service GPTZero is a offering for education professionals, while OpenAI is working on a "statistical watermark" that is applied during text creation, to prevent fraud potential.