New research published by the Centre for Economic Policy Research found that among 26,811 Chinese students in grades seven through 12, AI adoption increased homework scores by 18% and cut down completion time by 30%. However, within six months, monthly exam scores decreased by 20%, and college entrance examples fell by 18% to 24%—with scores reaching their worst after two years.

Researchers from Stockholm University and University of Hong Kong pinpointed the type of student most likely to experience these diverging scores: those who “outsourced” homework, deploying AI to complete the homework accurately, but in little time. The poor test scores, both in the long- and short-term were driven by about 80% of these students. The emergence of this student profile not only exposes a dissonance in AI productivity versus actual productivity gains—it fuels an argument some educators and experts have against the unfettered use of technology in education more broadly.

“For students, completing these tasks efficiently is not the goal; learning from them is,” researchers wrote. “Hence, the rapid diffusion of generative AI tools among students in recent years has created widespread concerns about their learning…Our findings show that generative AI, which is likely to become a prevalent technology for education, has a substantial negative impact on student learning.”

Gen Z’s interest or ability to engage critically with educational materials is being increasingly questioned as the generation becomes synonymous with AI adoption—and cheating at school. An Atlantic cover story has hinted at nothing less than the onset of a new Dark Ages, fueled by a “post-literate” younger demographic concerned more with quickly and conveniently intaking huge amounts of information, and less with savoring and digesting it, atrophying the ability to think critically. This research suggests that whatever the reason, the incentives to use AI to skip the act of learning are simply overwhelmingly powerful—and a problem society is failing to grapple with.

Why students turn to AI

AI use in schools has proliferated globally, with 84% of U.S. high schools students reporting using the technology for homework, according to a CollegeBoard survey of more than 1,000 high schoolers. Along with greater adoption has come misuse of the technology. Jacob Shelley, an associate professor of health law at Western University, told Fortune in May he was convinced his students cheated on a final exam, including using AI, for one of his classes, with 8% getting a perfect school on the multiple choice section, only to struggle on the essay portion, submitting answers with content not in the curriculum. 

“The results were anomalous,” Shelley said. “That just never happened in 20 years of teaching.” 

But rather than blame students for turning to the technology in high stakes moments, Shelley said he understands why students would feel compelled to cheat. Tech leaders like Anthropic’s Dario Amodei and OpenAI’s Sam Altman are now walking back predictions of an AI job apocalypse, but anxiety around the future of work in the world of AI still lingers. Computer scientist Cal Newport called these premonitions “doom trolling,” accusing tech companies of manufacturing a fatalistic narrative around AI. They appear to have had an impact on the generation preparing to enter the workforce: Almost 90% of graduates from the class of 2026 are worried AI or automation could replace entry-level jobs, according to job search platform Monster.

While economic data has yet to show an impact from AI on the labor market or productivity, Shelley said his students still feel the pressure to use the technology or risk being left behind.

“AI is going to replace them, at least a lot of them, and they know that, and we’re pretending that it won’t,” he said. “I think they see through it. So students are responsible, but I don’t really blame them here.”

The folly of the teaching machine

It may be no surprise to experts like neuroscientist Jared Cooney Horvath why homework gains thanks to AI aren’t translating to learning or exam performance. Horvath—who wrote in a testimony to the U.S. Senate Committee on Commerce, Science, and Transportation about how test scores indicate Gen Z is the first generation to be less cognitively capable than their parents—has long opposed educational technology, or EdTech. He argues there’s more than 100 years of evidence indicating automation can hinder learning, beginning in 1924 with the invention of the “teaching machine” Ohio State University psychology professor Sidney Pressey. Students would answer questions that a machine would displace when fed a piece of paper, but when asked outside the device to generalize their knowledge, they were unable to.

Three decades later, legendary behaviorist B.F. Skinner produced his own version of the machine based on Pressey’s prototype, where students would press keys indicating the correct answer, at which point another question would appear. But despite more advanced technology behind the mechanism, it yielded the same results, leaving both psychologists to abandon the project before it was implemented in schools. In a letter to Skinner, Pressey conceded that while students had not mastered the subject matter; they had just mastered the machine.

“The reason they all quit was the transfer problem,” Horvath said. “They found that kids would be very good so long as they were using the tool, but as soon as they went off the tool, they couldn’t do it anymore.”

AI learned has the potential to once again recreate the problems of the teaching machine, Horvath argued. While teachers have found some benefits to AI in the classroom—such as scaffolding text to individual students’ lexile levels, particularly English-language-learners—Horvath has deja vu. AI can individualize learning by generating answers to specific queries, but it does not produce the friction or enable the critical thinning necessary for learning subject matters, he argued.

“The tools experts use to make their lives easier are not the tools children should use to learn how to become experts,” Horvath said. “When you use offloading tools that experts use to make their lives easier as a novice, as a student, you don’t learn the skill. You simply learn dependency.”