Faster solutions, lower test scores: How AI is eroding math skills

AI's Shadow: Are Students Trading Speed for Smarts in Math?

The arrival of advanced AI tools in late 2022 sparked immediate questions among educators: would students leverage this technology for cheating, genuine learning, or simply a faster path through homework? Emerging evidence suggests a troubling trend: many students are completing assignments with unprecedented speed, but at the cost of deeper understanding.

This stark conclusion stems from one of the most extensive analyses to date on how generative AI is reshaping student behaviors and academic proficiencies. Sina Rismanchian, a doctoral candidate at the University of California, Irvine, collaborated with researchers from a prominent educational technology firm to scrutinize millions of student interactions with an online math platform. This platform, utilized by over four million students annually from middle school through college, offers a unique vantage point.

Because the platform encompasses both low-stakes practice exercises and high-stakes college placement assessments, the research team could meticulously compare student performance and engagement patterns before and after the widespread availability of AI chatbots.

Unpacking the AI Effect: A Tale of Two Problem Types

To isolate the specific impact of AI, the researchers focused on two distinct categories of math problems: word problems and graphing problems. These categories were chosen because of their varying susceptibility to AI assistance. Word problems, with their textual nature, can be easily copied and pasted into AI interfaces for immediate solutions.

Graphing problems, conversely, present a far greater hurdle for AI outsourcing. A student would need to capture a screenshot of the problem and then meticulously recreate the graph within the platform's specific tools, a process far less conducive to quick AI intervention.

Following the public debut of sophisticated AI, a clear divergence in student engagement and performance emerged between these two problem types. Beginning in early 2023, a noticeable decrease in the time students dedicated to word problems became apparent, while their engagement with graphing problems remained relatively stable.

The Widening Gap: Time Spent on Math Shrinks

This disparity in time investment continued to grow with each subsequent academic quarter. By the study's conclusion, nearing the end of 2025, the average time spent on word problems had plummeted significantly. High school students saw a 31 percent reduction, while college students experienced a 27 percent decline.

This translates to a drop from approximately four minutes per word problem to under three minutes. While younger students, specifically fifth graders, showed minimal impact, middle school students exhibited a modest decline of 9 percent. Researchers posit that these overall averages are being skewed downwards by a subset of students who are spending mere seconds on word problems, strongly indicating the use of AI for instant answers.

A parallel trend was observed in college placement tests. When these exams were administered in unsupervised settings, students dedicated substantially less time to word problems after the advent of AI. However, during proctored exams, the time spent on these problems reverted to historical norms, suggesting a reliance on AI when not under direct observation.

Beyond Time: The Erosion of Understanding

But time is only one facet of this complex issue. The more concerning finding lies in what happened to actual learning. Many higher education institutions permit incoming students to retake placement tests after engaging in further math practice on platforms like the one studied, offering a pathway to qualify for more advanced courses.

Historically, this additional practice yielded tangible improvements in student outcomes. Post-AI, however, a disturbing pattern emerged: students were answering more word problems correctly during unsupervised practice sessions, only to perform significantly worse on the same types of problems when later taking a supervised placement test.

Prior to the widespread adoption of AI, students typically answered around 80 percent of these word problems correctly on supervised placement tests. Following the introduction of AI, this figure dropped to approximately 60 percent, representing a substantial 25 percent decrease in the likelihood of correctly solving a word problem.

In stark contrast, performance on graphing problems remained unaffected. This stability suggests that the observed decline in word problem proficiency is not a symptom of a general deterioration of math skills due to factors like pandemic-related learning loss, weaker prior academic preparation, or pervasive digital distractions.

The Unseen Hand: Inferring AI's Influence

While the study cannot definitively prove that students were actively using AI for every instance of reduced time or performance decline, the evidence strongly points in that direction. The researchers were unable to monitor activities outside the specific platform, but the observed patterns are difficult to attribute to other causes.

The changes were exclusively seen in problems easily solvable by AI, vanished under direct supervision, and consistently grew over a nearly three-year period. This convergence of factors paints a compelling picture of AI's pervasive influence on student study habits and academic outcomes.

"What makes me nervous is that it's not only about the word problems," Rismanchian shared. "This cognitive surrender might be going on in writing, science, everything." The research paper, titled "Faster Completion, Less Learning," was released as a working paper and is undergoing peer review, acknowledging that a single study rarely settles complex questions.

A Growing Body of Evidence: The "Cognitive Surrender"

This investigation joins an expanding collection of research suggesting that generative AI is prompting students to bypass the essential cognitive effort that underpins genuine learning. This phenomenon, termed "cognitive surrender," appears to be becoming increasingly prevalent across various academic disciplines.

Further supporting this concern, a randomized experiment conducted in another country found that high school students who utilized AI for math study ultimately acquired less knowledge than their peers who practiced without such assistance. Independent reports from developers of AI language models also indicate that a significant number of college students are employing AI to obtain answers and offload cognitive tasks.

Rismanchian's prior research, published earlier in 2026, highlighted concerning patterns of AI usage in short-response essays among undergraduate students at a major research university, further underscoring the broad applicability of these trends.

AI as a Tool, Not a Crutch: Redefining Engagement

This does not imply that AI invariably undermines learning. When thoughtfully designed, AI-powered tutors have demonstrated the capacity to enhance student achievement in controlled experimental settings. These effective AI tutors function by posing targeted questions, personalizing instruction, and strategically withholding answers until students have thoroughly reasoned through a problem.

Crucially, this type of AI engagement is intended to *increase* the time students spend grappling with a problem, according to Rismanchian. The data from the ALEKS platform, however, shows the opposite trend – a significant reduction in time spent, particularly on AI-susceptible tasks.

Rismanchian does not advocate for a complete ban on AI tools. Instead, he emphasizes the need for students to cultivate a profound appreciation for the learning process itself, empowering them to resist the allure of outsourcing their intellectual efforts. This internal motivation is key to navigating the challenges posed by readily available AI assistance.

Students' Own Concerns: Acknowledging the Threat

A recent survey by a national research organization reveals that many students are already acutely aware of the potential threat AI poses to their cognitive abilities. A substantial portion of students report concerns that AI is weakening their critical-thinking skills, even as more of them admit to using it for their academic work.

This complex situation is not solely attributable to student choices. While many instructors have issued warnings against using AI for coursework, educational institutions themselves have often embraced the technology, sometimes providing students with free access to premium AI chatbot services. This institutional stance can inadvertently create a mixed message for students.

"I think we need to communicate to students that you should value your learning," Rismanchian stated. "If ChatGPT does it for you, then you haven't learned it." This direct message underscores the fundamental principle that true learning requires active engagement and personal effort.

A Personal Reckoning: The Author's Own Experience

Rismanchian himself understands the powerful temptation that AI presents. As an international student, he initially turned to AI to refine the English in his academic papers, ensuring his ideas remained his own while improving the clarity of his writing.

However, after several months of this practice, he experienced a disquieting realization. "I realized that I cannot write anymore," he recounted. "I was losing my writing abilities." This personal encounter with the potential for skill degradation led him to cease using AI for writing tasks, though he continues to utilize it for coding, recognizing its utility in specific domains.

The implications of this research extend far beyond mathematics, raising critical questions about the future of learning in an era of pervasive AI. As students navigate these new technological landscapes, the balance between efficient completion and deep, lasting learning will remain a central challenge for educators and students alike.