How capable is GPT-4 at answering exams and tests in Norwegian, and what implications could this have for education?

A number of meta-analyses have compared traditional teaching and tutoring with Intelligent Tutoring Systems (ITSs) and have discovered that ITSs can perform well in one-on-one tutoring for students. However, the existing ITSs have not handled the Norwegian language, and although large language models like GPT-4 are not developed as ITSs, they have clear similarities. The purpose of this position paper is to address how GPT-4 handles multilinguistic and cognitive challenges with Norwegian as an exam/test language, descriptive-procedural questions, and various exam and test contexts. This includes exams in medicine, nursing, psychology, dentistry, a theory test for military service, the Norwegian driving test, the Swedish university entrance test, the Norwegian citizenship test, and a national teacher exam in mathematics. A key concern is to investigate what kind of implications this could have for education. The results show that GPT-4 handles multilinguistic challenges well and performs remarkably well in Norwegian within both academic exam tasks and generic IQ and social tests, with elements of both descriptive and procedural questions. With an average precision rate of 94.26%, GPT-4’s performance demonstrates a high degree of reliability. This is partly consistent with the international knowledge base in some areas, but the study differs from this by confirming that it also masters Norwegian exam and test contexts very well. Thus, it has the ability to adapt linguistically and culturally, something conventional ITSs do not possess. GPT-4 (Spring 2024) is also capable of conducting multimodal analyses of medical images in Norwegian, and handles minority languages in Sami, which ITSs usually cannot do. Despite certain limitations, the position paper claims that such findings contribute to a broader understanding GPT-4’s multilinguistic and cognitive capabilities in formative and summative assessment contexts. Therefore, this position paper argues that it has significant implications for various aspects of the educational sector.