Students in Brazil say it is ethical to use ChatGPT in tasks – 08/24/2023 – Education

Brazilian students consider that using ChatGPT to do study assignments is ethical and should be accepted. They also claim that they would use the artificial intelligence tool to carry out work in the next semester.

This is what was reported to the authors of a survey published this Thursday (24th) by the journal Scientifc Reports.

The scientists sought to assess how students and educators evaluate the use of ChatGPT. In addition to Brazil, India, Japan, the United Kingdom and the United States participated in the study, with 100 educators and 200 students interviewed per country.

The result seen in Brazil is not something out of the norm. According to the researchers, there was a general consensus among educators and students that the use of ChatGPT in schools should be recognized and that it could increase the competitiveness of students who are not native speakers of English.

When looking only at students, a consensus was seen that in their future jobs there will be the possibility to outsource simple tasks to ChatGPT, which supposedly could allow them to focus on more creative and substantial parts of their work. .

According to the study, 94% of Brazilian students interviewed said they would use the robot for work in the next semester.

With the data available, the researchers ran a statistical regression to try to find patterns about students using or not using ChatGPT. Scientists point out that Brazilian and Indian students are significantly more likely to use than US students. Meanwhile, those in Japan are less.

Students from poorer and working-class backgrounds are significantly more likely to use ChatGPT than those from higher classes, the survey finds.

Although a consensus has been seen in the recognition of the use of ChatGPT, opinions vary on whether or not it is ethical to use it in school work. In the USA and India, students saw such use as wrong and were in favor of the ban. In Brazil, the opposite was observed.

India was the only country where students and educators said they were concerned about ChatGPT taking their jobs in the future.

Looking specifically at educators, respondents in Brazil and Japan say ChatGPT will reduce inequality in education. The other countries do not agree with this idea.

Scientists also compared robot and student performance

The scientists also tested in the study how ChatGPT would fare against students across 32 college courses, in a style of performance testing that has quickly become classic when thinking about models of natural language processing.

In 9 of the 32 courses, the chatbot result was comparable and even surpassed human performance. This was for Data Structures, Introduction to Public Policy, Object Oriented Programming, and six other courses.

Mathematics and questions with “gotchas” were the weak point of ChatGPT, in comparison to the performance of the students —as even showed a test done by Sheet on performance in the exam. “For now, humans appear to outperform ChatGPT in these areas,” the authors wrote.

Interestingly, the difference in performance between the robot and humans is significantly smaller for questions that require high levels of knowledge and cognitive processing.

The comparison between the chatbot and humans was possible because the scientists asked educators ten questions that had been used in university courses and three answers, for each question, given by randomly chosen students.

The researchers then generated with ChatGPT three responses to each of the ten questions. All responses, human and robotic, were randomly placed into a document. Three educators then graded the responses.

Finally, the researchers tried to see if other artificial intelligence tools were capable of detecting responses given by language models and by humans. For this, they used GPTZero35 and a program from OpenAI itself, developer of ChatGPT.

The OpenAI application wrongly classified 5% of human submissions as being created by a chatbot. In a kind of self-deception, he classified 49% of the responses made by ChatGPT as being produced by humans. For GPTZero, the numbers were respectively 18% and 32%.