How does ChatGPT affect the design of exams in higher education?
With the release of ChatGPT, the higher education world is in trouble. Almost overnight, an established cultural practice is reduced ad absurdum. Essays, case studies, or theses appear to be obsolete for measuring the acquisition of competencies.
Disruptive changes have been looming for some time. As universities, we must take a stand and consider the consequences of curricula and examinations. Then again, we should also take the time for a well-founded discussion.
Universities will not be able to ban the use of AI in the long term.
The challenge is evident: A ban on using AI will not be enforceable. First attempts at identifying whether a text was written by an AI or not have been made by analyzing language patterns. Well then? Does an exam performance with a 90% identified AI result in a lower grade?
Also, we do not have the exact criteria to evaluate available AI tools.
1) languagetool.org offers a grammar, style, and spelling checker that can be integrated into browsers, e-mail programs, and word processors. Apart from correcting errors, the tool also provides assistance in revising phrases and points out stylistic improvements.
2) Deepl.com is a text translation platform. The outcome of this translation tool always amazes me. The option of using an individual glossary to feed the tool with “technical terms” also makes the tool suitable for supporting the translation of specialized texts.
In terms of technology, they are both based on complex language models that are used to derive decisions. Deepl-Write, the AI writing assistant from Deepl, therefore does not appear to be a giant leap.
How should we evaluate the “degree of intelligence” of a tool? Is AI that revises my text more “forbidden” than AI that suggests a text for me to revise? How intelligent does a tool have to be for it to be forbidden? Will there eventually be a constantly maintained blacklist of forbidden tools?