When ChatGPT showed up in late 2022, educators quickly asked whether trainees would use artificial intelligence to cheat, find out or merely make it through homework more efficiently. Proof is starting to point toward an uncomfortable response: Lots of students appear to be finishing tasks quicker while learning less from them.
This conclusion originates from among the biggest studies of how generative AI is changing trainee habits and scholastic skills. Sina Rismanchian, a doctoral trainee at the University of California, Irvine, partnered with scientists at McGraw Hill to analyze millions of student interactions with ALEKS, an online mathematics platform utilized by more than 4 million students a year, from fifth grade through college. Due to the fact that ALEKS consists of both low-stakes practice problems and college placement tests, the scientists had the ability to compare how trainees acted and carried out previously and after ChatGPT’s arrival.
To separate AI’s effects, the researchers compared two kinds of math problems that vary in how quickly trainees can outsource them to AI: word issues and graphing problems.
Word issues can be copied and pasted straight into AI chatbots for immediate answers. Graphing problems are far more cumbersome. A trainee would need to upload a screenshot and still recreate the chart inside ALEKS utilizing its tools.
After ChatGPT’s launch, student behavior and efficiency on the two kinds of issues started to diverge.
Starting in early 2023, trainees began spending less time on word issues while continuing to spend about the very same quantity of time on graphing problems. The gap expanded every quarter. By the end of the study period, near completion of 2025, typical time spent on word issues had fallen 31 percent amongst high school trainees and 27 percent amongst university student– from about 4 minutes per word issue to less than 3. (Middle school students showed just a modest decline of 9 percent, and fifth graders revealed essentially none.)
The researchers think those averages are being pulled downward by some students who invest only seconds on word problems because they’re using AI to address them.
The same pattern appeared in college positioning tests. When the tests were taken without guidance, students spent much less time on word issues after ChatGPT’s release. Throughout proctored exams, the time invested in word issues returned to historic standards.
But time is only half the story. The more troubling finding is what occurred to finding out.
Many colleges allow inbound students to retake positioning tests after practicing more mathematics in ALEKS, giving them a possibility to qualify for a higher-level course. Before ChatGPT, that practice typically settled. After ChatGPT, students addressed more word issues correctly throughout without supervision practice sessions but performed significantly worse on those exact same kinds of problems when they later on took a proctored positioning test.
Historically, students addressed about 80 percent of these word issues properly on supervised placement tests. After ChatGPT’s intro, that was up to about 60 percent– an approximately 25 percent decrease in the chances of responding to a word problem correctly.
Efficiency on graphing issues, by contrast, did not decline.
After ChatGPT’s release, trainees performed worse on word issues (AI-susceptible) during proctored tests, but answer more word issues correctly in nonproctored settings
The dotted line marks the public release of ChatGPT. Source: Figure 4, Rismanchian et al “Faster Completion, Less Knowing: Generative AI Minimized Study Time on Math Problems and the Understanding They Construct,” June 2026 preprint.
If students’ mathematics abilities had actually generally weakened because of pandemic knowing loss, weaker high school preparation or digital distraction, graphing performance ought to have deteriorated too. It didn’t.
The research study can not definitively show that trainees were utilizing AI. The researchers could not see what else was happening on trainees’ screens beyond ALEKS. However it’s challenging to think of another explanation. The changes appeared only in problems that are simple to contract out to AI, disappeared under guidance and grew progressively over almost 3 years.
“What makes me nervous is that it’s not just about the word problems,” Rismanchian informed me. “This cognitive surrender might be going on in writing, science, everything.”
The paper, “Faster Conclusion, Less Learning,” was launched in June 2026 as a working paper and has not yet been peer evaluated. Like any single research study, it does not settle the concerns of how much students are utilizing AI in their schoolwork, whether it’s damaging learning and by how much. But it signs up with a growing body of proof that generative AI is causing students to skip the brain work that results in discovering, and that this “cognitive surrender” is ending up being commonplace.
A randomized experiment in Turkey discovered that high school trainees who utilized AI to help them study mathematics ultimately found out less than students who practiced without it. Anthropic, the maker of Claude, has actually individually reported that lots of university student appear to use AI to obtain answers and unload cognitive work. Rismanchian’s earlier research study, released in March 2026, recorded unpleasant patterns of AI use simply put reaction essays among undergraduate students at a large California research university.
That does not imply AI constantly weakens learning. Carefully designed AI tutors have enhanced trainee accomplishment in regulated experiments by asking questions, individualizing guideline and withholding responses until trainees reason their way through an issue. But using AI by doing this must increase the time students invest in a problem, Rismanchian said. The ALEKS data show the opposite.
Rismanchian does not believe the response is just prohibiting AI. Rather, he argues, students need to worth learning enough to resist the temptation to outsource it.
A current RAND survey suggests numerous already acknowledge the risk to their brains. Trainees report fretting that AI is compromising their critical-thinking skills while more of them admit using it for schoolwork.
Students are not entirely to blame. Even as many teachers have actually warned trainees not to use AI to finish classwork, universities themselves have actually accepted the technology, typically providing students free access to premium chatbots.
“I think we need to interact to trainees that you ought to value your learning,” Rismanchian said. “If ChatGPT does it for you, then you have not discovered it.”
Rismanchian understands the temptation.
An international trainee, Rismanchian began using ChatGPT to assist polish the English in his papers. The concepts were still his own. But after numerous months, he said, he saw something unsettling.
“I understood that I can not write any longer,” he said. “I was losing my writing abilities.”
So he stopped using AI to write.
He still uses it to code.
Contact staffauthor Jill Barshay at 212-678-3595, jillbarshay.35 on Signal, or [email protected].
This story about AI use wearing down mathematics skills was produced by The Hechinger Report, a nonprofit, independent wire service that covers education. Register for Proof Pointsand other Hechinger newsletters.
Was this story handy? Leave a pointer to support your education reporters.
The Hechinger Report is a not-for-profit newsroom powered by reader support
Republish our short articles for free, online or in print, under an Imaginative Commons license.