The rapid advancement of artificial intelligence (AI), particularly generative tools like ChatGPT, presents emerging challenges for academic integrity in higher education. This study evaluates ChatGPT 3.5′s performance on 500 instructor-generated exam questions from undergraduate criminal justice courses, encompassing multiple choice, true/false, and short answer formats. Results indicated an overall accuracy rate of 80%, with notable variation across question types and individual chatbot instances. ChatGPT answered true/false questions with the highest accuracy (83%) and short answer questions with the lowest (72%). Although responses were generally accurate, consistency across test accounts was limited, suggesting that outcomes from AI-assisted exam-taking may differ between users. Additionally, restrictions on copying and pasting were found to significantly reduce response speed, highlighting a practical deterrent strategy. These findings underscore the need for reconsidering assessment design in light of AI capabilities and offer recommendations for mitigating AI-facilitated academic dishonesty in online-based evaluations.
Loading...
Cundiff et al. (2025) studied this question.