Scholarlink

<p>The rise of generative artificial intelligence (AI) poses major challenges for online assessment in higher education. This study examines how the unauthorised use of AI tools affects exam outcomes by comparing six exam sessions (three online, three in-person) with more than 600 students. Findings show that online exams without supervision yielded unusually high average grades and very short completion times, indicating AI-assisted cheating. In contrast, supervised in-person exams reflected lower but more realistic performance. A key methodological intervention was the semantic modification of exam questions, where more complex phrasing reduced the advantage of AI use and narrowed the performance gap between online and in-person formats. Statistical analyses (ANOVA, Kruskal–Wallis, post-hoc tests) confirmed that exam format and question design explained a large proportion of grade variance (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\eta ^2 = 0.667\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mi>η</mi> <mn>2</mn> </msup> <mo>=</mo> <mn>0.667</mn> </mrow> </math></EquationSource> </InlineEquation>). The study concludes that while AI misuse threatens exam integrity and fairness, targeted interventions—such as linguistic reformulation—can mitigate its impact. These insights provide practical recommendations for universities seeking to adapt their assessment cultures in the AI era.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The Educational and Assessment Implications of Unauthorised Use of Artificial Intelligence in Online Examinations: A Quantitative Analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月19日

查看全文
  • Tamás Klein,
  • József Zoltán Málik

摘要

The rise of generative artificial intelligence (AI) poses major challenges for online assessment in higher education. This study examines how the unauthorised use of AI tools affects exam outcomes by comparing six exam sessions (three online, three in-person) with more than 600 students. Findings show that online exams without supervision yielded unusually high average grades and very short completion times, indicating AI-assisted cheating. In contrast, supervised in-person exams reflected lower but more realistic performance. A key methodological intervention was the semantic modification of exam questions, where more complex phrasing reduced the advantage of AI use and narrowed the performance gap between online and in-person formats. Statistical analyses (ANOVA, Kruskal–Wallis, post-hoc tests) confirmed that exam format and question design explained a large proportion of grade variance ( \(\eta ^2 = 0.667\) η 2 = 0.667 ). The study concludes that while AI misuse threatens exam integrity and fairness, targeted interventions—such as linguistic reformulation—can mitigate its impact. These insights provide practical recommendations for universities seeking to adapt their assessment cultures in the AI era.