ChatGPT Excels at University Assignments According to Study

A recent study published in Scientific Reports reveals that ChatGPT, an artificial intelligence language model, performs exceptionally well when answering university assessment questions. The researchers found that ChatGPT’s answers matched or even surpassed the average grades achieved by students in various subjects, including computer science, political studies, engineering, and psychology.

Researchers Talal Rahwan and Yasir Zaki conducted the study by inviting faculty members from New York University Abu Dhabi to provide three student submissions for 10 assessment questions. ChatGPT was then tasked with generating answers to these questions, which were evaluated alongside the responses from students by three graders. In nine out of 32 courses, ChatGPT’s answers received an average grade at least equal to or higher than the students’ grades.

The study revealed that ChatGPT outperformed students significantly in the “Introduction to Public Policy” course, with an average grade of 9.56 compared to the students’ 4.39. However, mathematics and economics courses were the only subjects where students consistently outperformed ChatGPT.

Additionally, the study surveyed 1,601 individuals from Brazil, India, Japan, the US, and the UK to gauge their views on utilizing ChatGPT for university assignments. Surprisingly, 74% of the students expressed their willingness to use ChatGPT as a tool for their work. Conversely, educators across all countries underestimated the number of students who planned to use ChatGPT, with 70% of them considering its use as plagiarism.

Furthermore, the researchers discovered that existing tools for identifying AI-generated text misclassified ChatGPT’s answers as human-written a significant portion of the time.

These findings shed light on the potential use of AI tools in educational settings and could inform policies regarding their incorporation. As the capabilities of language models like ChatGPT continue to improve, their role in assisting students with assignments could become more prevalent, despite concerns regarding plagiarism.

Source:
Talal Rahwan, Perception, performance, and detectability of conversational artificial intelligence across 32 university courses, Scientific Reports (2023). DOI: 10.1038/s41598-023-38964-3. (Nature Publishing Group)