‘Cognitive Surrender’: Faster Solutions, Lower Test Scores Show How AI is Eroding Math Skills

Beginning in early 2023, students started spending less time on word problems while continuing to spend about the same amount of time on graphing problems. The gap widened every quarter. By the end of the study period, near the end of 2025, average time spent on word problems had fallen 31 percent among high school students and 27 percent among college students — from about four minutes per word problem to less than three. (Middle school students showed only a modest decline of 9 percent, and fifth graders showed essentially none.)

The researchers believe those averages are being pulled downward by some students who spend only seconds on word problems because they’re using AI to answer them.

The same pattern appeared in college placement tests. When the exams were taken without supervision, students spent much less time on word problems after ChatGPT’s release. During proctored exams, the time spent on word problems returned to historical norms.

But time is only half the story. The more troubling finding is what happened to learning.

Many colleges allow incoming students to retake placement tests after practicing more math in ALEKS, giving them a chance to qualify for a higher-level course. Before ChatGPT, that practice generally paid off. After ChatGPT, students answered more word problems correctly during unsupervised practice sessions but performed substantially worse on those same kinds of problems when they later took a proctored placement test.

Historically, students answered about 80 percent of these word problems correctly on supervised placement tests. After ChatGPT’s introduction, that fell to about 60 percent — a roughly 25 percent reduction in the odds of answering a word problem correctly.

Performance on graphing problems, by contrast, did not decline.

After ChatGPT’s release, students performed worse on word problems (AI-susceptible) during proctored exams, but answer more word problems correctly in nonproctored settings

If students’ math skills had generally deteriorated because of pandemic learning loss, weaker high school preparation or digital distraction, graphing performance should have deteriorated too. It didn’t.

The study cannot definitively prove that students were using AI. The researchers couldn’t see what else was happening on students’ screens outside of ALEKS. But it’s difficult to think of another explanation. The changes appeared only in problems that are easy to outsource to AI, disappeared under supervision and grew steadily over nearly three years.

“What makes me nervous is that it’s not only about the word problems,” Rismanchian told me. “This cognitive surrender might be going on in writing, science, everything.”

The paper, “Faster Completion, Less Learning,” was released in June 2026 as a working paper and has not yet been peer reviewed. Like any single study, it doesn’t settle the questions of how much students are using AI in their schoolwork, whether it’s harming learning and by how much. But it joins a growing body of evidence that generative AI is causing students to skip the brain work that leads to learning, and that this “cognitive surrender” is becoming commonplace.

A randomized experiment in Turkey found that high school students who used AI to help them study math ultimately learned less than students who practiced without it. Anthropic, the maker of Claude, has separately reported that many college students appear to use AI to obtain answers and offload cognitive work. Rismanchian’s earlier research, released in March 2026, documented troubling patterns of AI usage in short response essays among undergraduate students at a large California research university.

That doesn’t mean AI always undermines learning. Carefully designed AI tutors have improved student achievement in controlled experiments by asking questions, personalizing instruction and withholding answers until students reason their way through a problem. But using AI this way should increase the time students spend on a problem, Rismanchian said. The ALEKS data show the opposite.

Rismanchian doesn’t believe the answer is simply banning AI. Instead, he argues, students need to value learning enough to resist the temptation to outsource it.

A recent RAND survey suggests many already recognize the threat to their brains. Students report worrying that AI is weakening their critical-thinking skills while more of them admit using it for schoolwork.

Students are not entirely to blame. Even as many professors have warned students not to use AI to complete classwork, universities themselves have embraced the technology, often giving students free access to premium chatbots.

“I think we need to communicate to students that you should value your learning,” Rismanchian said. “If ChatGPT does it for you, then you haven’t learned it.”

Rismanchian understands the temptation.

An international student, Rismanchian began using ChatGPT to help polish the English in his papers. The ideas were still his own. But after several months, he said, he noticed something unsettling.

“I realized that I cannot write anymore,” he said. “I was losing my writing abilities.”

So he stopped using AI to write.

He still uses it to code.