ChatGPT Outperforms Students in Most Subjects, Except Math, Study Finds
In a recent study conducted by researchers at New York University Abu Dhabi, OpenAI’s ChatGPT has been found to outperform university students in various subjects, except for mathematics. The study, published in Scientific Reports, reveals that ChatGPT’s performance even surpasses that of the average student in computer science, political studies, engineering, and psychology.
The research team, led by associate professors Talal Rahwan and Yasir Zaki, collected questions from different courses taught at the university. Faculty members provided ten questions along with three randomly selected student answers for each question. Using ChatGPT, three distinct answers were generated for each of the provided questions.
To evaluate the performance of ChatGPT in comparison to students, the compiled answers from both sources were presented to assessors who were unaware of their origin. Surprisingly, the study found that ChatGPT’s performance was comparable or even superior to that of students in nine out of thirty-two courses. Additionally, the research highlighted that existing detection algorithms often misclassify human answers as AI-generated and vice versa.
However, the study also discovered that ChatGPT’s performance was notably weaker in math-related questions, followed by trick questions. For the time being, humans seem to excel in these areas. It is worth noting that using an obfuscation attack to evade AI filters resulted in a 95 percent failure rate of detection algorithms in identifying ChatGPT answers.
In addition to analyzing the performance of ChatGPT, the researchers surveyed students and staff from Brazil, India, Japan, the UK, and the US to gauge their perspectives on AI in educational environments. The results revealed a divide in opinions. While 57 percent of students expressed their intention to use ChatGPT for assistance with assignments, 64 percent expected their peers to do the same. On the other hand, 69 percent of professors stated that they would consider the utilization of ChatGPT as plagiarism.
Interestingly, the study found a consensus among educators and students regarding the acknowledgment of ChatGPT’s use in school work, particularly in increasing the competitiveness of non-native English speakers.
These findings have significant implications for the integration of AI tools in education. As ChatGPT proves its prowess in various subjects, policymakers and educators may need to revise their approach to ensure that students are adequately prepared for the future. While the study shows promising results for ChatGPT’s capabilities, it also underscores the importance of human expertise, particularly in complex mathematical problem-solving.
As AI continues to advance, it is clear that the relationship between humans and technology in educational settings will require careful consideration and robust guidelines to strike a balance between leveraging the benefits of AI and preserving the value of human intellect.
Overall, this study sheds light on the strengths and limitations of ChatGPT, prompting informed discussions on how to incorporate AI tools effectively in education while upholding academic integrity and providing equal opportunities for all learners.
