The Future of Education
Challenges for Computer Science Education Arising from new AI Systems like ChatGPT
Robert Mischak, University of Applied Sciences FH Joanneum (Austria)
Robert Pucher, University of Applied Sciences Technikum Wien (Austria)
Abstract
New intelligent AI-based technologies have the potential to limit severely the effectiveness and significance of familiar didactic elements, such as knowledge testing by means of seminar papers or the submission of source code as an exercise. This is the case when students are supported in their work by AI applications such as ChatGPT. ChatGPT produces astonishing and high-quality solutions in a matter of seconds - without any significant input from the students themselves. It is then often very difficult for the teachers to recognize whether the work was done by the student himself or by a machine. This phenomenon is not limited to computer science students, but it can be assumed that they are even more affected because of their relevant technical competencies.
ChatGPT is based on an artificial neural network technology, called “transformer networks”. This technology allows not only to take natural language as an input, put also gives the answers in several languages in grammatically correct sentences which are structured in a way similar sentences written by humans are. The network was trained by using most data available in the entire internet with a cutoff date of September 2021. Therefore, many answers seem to reflect a deep and profound knowledge. Our first assessment of ChatGPT generated answers showed that in many cases these answers follow a specific pattern. This pattern might allow trained teachers to recognize the source. The challenge is it is very difficult to prove the answers do not come from the student himself, as the answer is generated and not stored in the internet. All standard tools used to check for plagiarism fail utterly to detect the source of the text.
Thus, it is necessary for teachers of computer science courses to develop new didactic concepts that allow a fair and appropriate performance assessment of the students or to integrate the potentials of ChatGPT into the lessons in a meaningful way.
Therefore, some general considerations as well as practical recommendations for teaching are given.
