AI Helps Students Complete Assignments, But Blunts Their Ability During Exams
Artificial Intelligence (AI) has become an integral part of the learning activities of students and undergraduates, with its usage becoming widespread in developed nations.
A survey by Chegg last year showed that 80% of students in developed countries use AI to assist their studies. This figure has surged in recent surveys, reaching 94% in the United Kingdom and 93% in Germany.
However, behind the convenience offered by AI, a serious issue has emerged. The technology that enables students to complete assignments more quickly may actually cause their ability to face examinations to decline.
Research conducted by David Stromberg from Stockholm University, alongside Victor Lei and Wu Yanhui from the University of Hong Kong, involving 27,000 students aged 12-18 in China, uncovered shocking facts. Approximately 80% of students admitted to using AI tools such as Doubao and DeepSeek. After six months, their homework grades increased by an average of 18% across various subjects.
Not only did grades improve, but the time taken to complete assignments also dropped drastically, from an average of 64 minutes to 45 minutes. Yet, when entering the examination hall, these advantages vanished.
Students using AI recorded exam scores 20% lower than those who did not use it. This finding challenges the assumption that high assignment grades always reflect learning proficiency. Before the AI era, homework grades could be used to estimate student performance in exams. Now, students with the highest assignment grades could potentially be the group most at risk of performing poorly during exams.
Should schools ban AI? The answer is not that simple. The decline in exam scores occurred primarily among students who used AI to accelerate task completion. Conversely, students who used AI but still allocated as much time to the work as those not using AI experienced only a slight decline in performance.
This highlights a crucial point: AI itself is not the problem; rather, it is how students use it. Students who maintained their exam performance did not merely ask chatbots for answers to copy and paste. Instead, they tended to use AI as a personal tutor, for instance, to explain difficult concepts, test understanding, or assist in solving specific problems.
These findings align with other research by Zara Contractor and Germán Reyes from Middlebury College. Undergraduates who studied new topics with the help of chatbots recorded higher test scores than those who studied without AI. This advantage even persisted when tests were repeated one week later.
This means that AI can indeed be an effective learning tool, but there is a clear limit. If AI is used to assist thinking, it can strengthen the learning process. If AI is used to replace the thinking process, the results could backfire.
Learners now face a new temptation: before they even attempt to think through a problem, they can immediately ask AI for the answer. This marks a new chapter in the challenges facing education. AI may make students more productive in completing assignments, but if every difficulty is immediately delegated to a machine, students may lose the very ability that is tested in the classroom: the ability to think independently.