The future of exams according to ChatGPT

 Artificial Intelligence (AI) is the current buzzword of digitization, and generative AI for creating texts (e.g. ChatGPT), images (e.g. DALL-E) produces the most memorable results for a wide audience. As is always the case with software innovation, the longer-standing development of middleware (OpenAI) with its data science algorithms has not received the same media response, simply because it is harder to understand and image. With generative AI, it's easier, and that starts the hype curve -- from inflated expectations to disappointed illusions.

For education, the capabilities of generative AI, especially in text creation, represent both risk and opportunity. The range of AI support in this regard is broad. Arguably, no one would object to assistance in creating an outline, an abstract, help with translation into English, or making sure a text is grammatically correct. We have problems with the fact that in the preparation of an examination performance (especially as a text: exam, term paper) the performance of the AI is so comprehensive that it eclipses the individual performance of the human examinee. We would like to measure and evaluate the individual performance of the examinee!

As long as we measure the performance of the human in a category that can also be done by the machine, we have only ourselves to blame. There is a reason that after the advent of the pocket calculator we no longer ask for mental arithmetic in school math exams. Both in the teaching of knowledge and in the measurement of performance, we have to be realistic about the fact that environmental conditions have changed and a new technology is available for the long term.

To make it as realistic and practical as possible, we prepare students in an exam for a real situation of a task solution under time pressure:

1980er: "Meier, please come quickly to the boss and explain why the booking was done this way. You have learned that, haven't you?". The logical form of examination for this is a "closed book" exam, in which you have to reproduce something you have learned by heart without any aids and with only paper and pencil. Why? Precisely because the practice later asks exactly the same questions as we ask in the exam.

2000er: "Meier, in two hours the presentation will be at the board and we need a quick compilation with the proposal and the reasoning. Fire up the Internet and find some sources." This is, of course, an "open book" retreat, where all the tools are available, but the solution must be aggregated, concisely summarized, and presented from a variety of relevant and irrelevant information. (If, beyond this compilation and with a little more time, all available information is sifted and an own idea is to be developed on this basis, we are at the examination performance of the written term paper or thesis. This is already "open book" anyway).

We can presumably assume that from 2023 onwards, it will no longer be just "the Internet that is switched on", but Mr. or Mrs. Meier will use a generative, text-creating AI such as ChatGPT to solve problems in the business reality of the future. Consequently, we will also have to adapt our university examinations accordingly. In any case, prohibiting AI will not lead us into the future.

Translated with DeepL