As the capabilities of AI have improved, the temptation for college students to cheat using AI has increased. According to a Berkeley study published in May, 9% of students who used AI admitted to using it to cheat. The study was conducted in the spring of 2024 and considering how quickly the technology has been adopted, coupled with the response bias inherent in asking students to admit to cheating, the current rate of AI cheating among undergraduates is likely much higher. As I lamented over this issue, I realized that the choice facing students about whether or not to cheat on a test with AI is a classic example of the Prisoner's Dilemma.
Most students, even those with a solid understanding of the course material, would probably be able to improve their test score, even marginally, by consulting AI. If a student's classmates all use AI to cheat, their higher test scores will skew the grading curve, leaving the student worse off if he does not follow suit. Assuming the professor is oblivious, cheating becomes the dominant strategy for all students because it will give them a better score (higher utility) than if they do not cheat, irrespective of the choices of their classmates. However, ideally the UVA Honor Code would act as a deterent so that no students would cheat, resulting in the Pareto-Efficient solution.
No comments:
Post a Comment