Will University Exams soon be Obsolete? An Analysis of ChatGPT on Assignments from Computer Science Modules Regarding Plagiarism and Correct Performance

This paper investigates the impact of ChatGPT, an autoregressive language model developed by OpenAI, on academic teaching, particularly in the context of exams and grading. The study analyzes the performance of ChatGPT in solving tasks from six modules of a Computer Science course and checks the generated answers for plagiarism. The results indicate that ChatGPT performs well in solving simple tasks, such as definitions, with low plagiarism rates. However, it struggles with open-ended questions, abstract tasks, or justifications, showing high plagiarism rates and more incorrect answers. The findings also suggest that individual students using ChatGPT for cheating might be difficult to detect through plagiarism scanners, but detection becomes more feasible when multiple students use it simultaneously. In conclusion, ChatGPT should be seen as a tool rather than a competitor, providing potential for shifting human focus towards more complex tasks and challenges. The academic community should consider discussing the ethical use of ChatGPT and adapt examination methodologies accordingly, rather than banning its use