Research Press Release
Scientific Reports
August 25, 2023
ChatGPT has published papers reporting that students scored at or above average when answering written questions in various college subjects (e.g., computer science, political science, engineering, and psychology). ,Scientific Reportswill be published in In this study, about three-quarters of the students surveyed said they would like to use ChatGPT to help them with their university assignments, whereas many instructors said they would use ChatGPT. It has also become clear that the company believes its use constitutes an act of plagiarism.
Talal Rahwan and Yasir Zaki conducted a total of 32 courses at New York University Abu Dhabi (NYUAD) to investigate how ChatGPT performs compared to students in answering university written questions. We called on the instructor to provide the 10 short-answer questions assigned to the students and the responses of three students to each question. Next, I let ChatGPT generate responses to these 10 questions (3 different responses for each question), and 3 graders (who were not told who created the responses) shared the responses between the students and the ChatGPT responses. evaluated. In 9 of the 32 courses, the average grade of responses generated by ChatGPT was equal to or higher than the average grade of student responses. The only subjects where students consistently outperformed ChatGPT were math and economics courses. The ChatGPT average score was significantly higher than the average student score for the “Introduction to Public Policy” course (9.56 vs. 4.39).
The authors also tested 1601 people in Brazil, India, Japan, the US, and the UK (including at least 200 students and 100 faculty members from each country) using ChatGPT in solving university assignments. We conducted an awareness survey on the theme of whether it is acceptable to do so. 74% of students said they would use ChatGPT when working on assignments for college. In contrast, instructors across all countries underestimated the proportion of students who said they intended to use ChatGPT, with 70% of instructors treating ChatGPT use as an act of plagiarism. I answered.
Furthermore, the authors verified two types of tools for identifying AI-generated text (GPTZero and AI text classifier), and found that the rate of misclassifying ChatGPT-generated answers as human-generated answers was , reported that GPTZero was 32% and AI text classifier was 49%.
Taken together, the findings provide a glimpse of a number of clues that may inform policy regarding the use of AI tools in education.
doi:10.1038/s41598-023-38964-3
“Highlights of Interest in Nature” is a translated version of a press release produced by Nature’s public affairs department. If you need more precise and detailed information, be sure to refer to the original article.
Return to “Highlights of interest” article list