Introduction: A New Era of Learning Challenges
With the launch of ChatGPT in late 2022, educators found themselves grappling with the implications of artificial intelligence on student performance. While some hoped it would enhance learning, emerging evidence suggests a more troubling trend: students are completing math assignments faster but retaining less knowledge.
Research Foundation: Analyzing Student Interactions
This finding stems from a comprehensive study spearheaded by Sina Rismanchian, a doctoral candidate at the University of California, Irvine, in collaboration with researchers from McGraw Hill. The investigation focused on how generative AI impacted over four million students using ALEKS, an online math platform, by analyzing millions of student interactions before and after the arrival of ChatGPT.
Dissecting Problem Types: Word vs. Graphing
To truly grasp AI's effects, the researchers categorized math problems into two distinct types: word problems, which are easily handled by AI chatbots, and graphing problems, which require more engagement on the part of students. Students could quickly copy and paste word problems into AI, while tackling graphing issues demanded a more hands-on approach, requiring screenshots and manual graphing.
Behavioral Changes Post-ChatGPT
After the introduction of ChatGPT, notable shifts in student behavior and performance emerged. Beginning in early 2023, students consistently devoted less time to word problems while maintaining their engagement with graphing problems. By the end of the study period in late 2025, average completion time for high school students on word problems dropped by 31%, and by 27% for college students. In stark contrast, younger students showed minimal change in their approach.
The Dichotomy of Performance: Testing Conditions Matter
The dichotomy in performance surfaced clearly during college placement tests, where students demonstrated a striking contrast in behavior based on whether the exams were supervised or not. In unsupervised settings, students exhibited a significant decrease in time spent on word problems, while their performance on supervised assessments returned to historical levels.
Learning Outcomes: A Decline in Proficiency
The most alarming finding wasn't just about time—it was about learning. In instances where students were given the opportunity to retake placement tests after practicing math through ALEKS, a worrying pattern emerged. Although practice sessions saw improved responses to word problems, the actual test results suffered. Historically, students answered around 80% of word problems accurately during supervised tests, but that accuracy plummeted to approximately 60% following the arrival of ChatGPT—a staggering 25% drop.
Understanding the Graphing Problem Performance
Interestingly, performance on graphing problems remained steady, suggesting that the issue doesn't stem from pandemic disruptions or a general decline in student ability. If such skills had deteriorated academically, one would expect similar declines across all problem types, yet that was not the case.
The AI Usage Quandary: Correlation or Causation?
The study does not conclusively prove that students were using AI extensively; the researchers could not monitor external screen activities. Yet, given that negative performance trends were evident only in problem types conducive to AI assistance, the correlation seems significant. Rismanchian emphasized, “What concerns me is that this could extend beyond math to writing, science, and beyond.”
Evidence from Other Studies
This study aligns with other findings, such as a randomized experiment in Turkey where high school students utilizing AI tools to study math learned less than peers who practiced independently. Furthermore, concerns about college students increasingly using AI for coursework had been echoed in reports by Anthropic, highlighting a widespread trend of relinquishing cognitive work.
Potential Benefits of AI in Education
While unequivocal negative impact was illuminated through this research, not all AI applications undermine learning. Some well-structured AI tutors have been shown to enhance student performance by providing tailored instruction. However, the data from ALEKS indicate that AI is currently pushing students to rush through problems rather than facilitating meaningful engagement.
Addressing the Challenge: Rethinking AI’s Role
Rismanchian advocates against outright banning AI in educational settings; instead, the focus should be on instilling a value of learning in students that prompts them to resist the allure of quick fixes. A recent RAND survey reflects this perspective, as many students themselves express concerns that AI undermines their critical-thinking skills, yet still report using it for academic tasks.
Institutional Responsibility: The Larger Picture
Students are not solely culpable in this situation. While educators caution against the over-reliance on AI for completing assignments, some universities aggressively promote these technologies, providing students at times with free access to premium AI services.
The Broader Implications: A Personal Reflection
Rismanchian shares his own experience with AI, initially using ChatGPT to refine his writing. However, he noticed a detrimental effect on his skills and subsequently ceased using the tool for that purpose, continuing only to apply it for coding tasks. His story echoes the broader implications of cognitive surrender across varying disciplines.
The findings raise critical questions about the direction of modern education in an era increasingly influenced by AI technologies. As educational institutions wrestle with how to integrate AI effectively, the call remains for a collective effort to prioritize learning that extends beyond expediency.
