Universities introduce oral exams to combat cheating with AI, while research highlights their inconsistency and unfairness

Universities, along with some schools in New South Wales, Australia, are beginning to make greater use of oral exams (“viva voce”) in response to students cheating with artificial intelligence. According to universities, this method allows them to assess a student's actual knowledge more accurately because the examiner can ask additional questions on the spot. Oral examination is among the oldest forms of assessment, dating back to ancient Greece, and was used in many schools and universities even before written examinations became widespread.

However, according to the cited research, the method has weaknesses of its own. A German study found that students experience significantly greater anxiety during oral exams than during written ones, which may prevent them from demonstrating their actual knowledge. Students examined in a language other than their native language may be disadvantaged even when they understand the material. A Norwegian study in secondary schools also showed substantial inconsistency: students' opening presentations lasted from five to 16 minutes, follow-up discussions from seven to 23 minutes, and in subsequent research, some students were asked fewer than ten questions by the same examiners, while others were asked almost fifty.

Oral exams are also unreliable as a defence against cheating with AI – a student can use AI to rehearse and memorise convincing answers, while another student who wrote their own work may fail under pressure. The article's author, whose own research focuses on assessment, states that effective assessment must produce trustworthy and reliable results, treat all students fairly, allow appropriate flexibility according to students' needs, and measure the intended learning – for example, scientific reasoning in an oral science exam, rather than confidence in public speaking.

According to the research cited in the article, oral exams work best as a complement to written or practical assessment, rather than as a replacement – for example, a short conversation about an annotated essay draft instead of replacing an entire essay with an exam. Recommended measures include communicating the purpose and assessment criteria in advance, opportunities to practise speaking regularly, enough time to think, adapting conditions to students, and consistent follow-up questions across students, supported by a shared marking rubric and a review of a sample of grades. Without these measures, the author argues, an oral exam used as an “AI lie detector” merely turns a test of knowledge into a test of confidence.