New FSU research uncovers a groundbreaking method to detect cheating using generative artificial intelligence on multiple-choice exams. This innovative study, conducted by chemist Ken Hanson from Florida State University, in collaboration with machine learning engineer Ben Sorenson, sheds light on how AI tools like ChatGPT can be identified through specific statistical analyses.
The research, published in the Journal of Chemical Education, marks a significant breakthrough in addressing the misuse of AI technology in academic settings. While concerns surrounding AI-driven cheating have predominantly focused on essays and narrative assignments, the use of tools like ChatGPT on multiple-choice exams has largely gone unnoticed until now.
Hanson and Sorenson’s study reveals that by comparing the performance of students on chemistry exams with those completed by ChatGPT, distinct patterns can be identified to detect instances of AI-based cheating. Through the application of statistical methods such as fit statistics and Rasch modeling, the researchers were able to pinpoint discrepancies in how AI-generated responses differed from those of human students.
The findings highlight the importance of implementing robust measures to combat cheating enabled by AI technology. By understanding the unique patterns exhibited by generative AI chat bots, educators can effectively identify and prevent their use in multiple-choice exams, ensuring academic integrity and fairness.
This groundbreaking research underscores the collaborative efforts between academia and technology to safeguard the integrity of educational assessments. As AI continues to play an increasingly prominent role in education, initiatives like Hanson and Sorenson’s study serve as a vital tool in maintaining academic honesty and ensuring that student evaluations remain a true reflection of their knowledge and abilities.