The new study followed 26,811 high school students in China for 30 months. The study measured how generative AI affected their learning.
Good news: Students using AI saw an 18% increase in homework scores and a nearly 30% decrease in completion time. Better scores are made faster. These are the results that every teacher should be happy to see.
Bad news: The same students saw their closed-ended test scores drop by 20% within six months. Entrance exam scores dropped from 18% to 24%, with the biggest drop after nearly two years of AI use.
About 80% of AI student users indicated outsourcing thinking results. They performed the tasks unusually fast. They got good marks on their homework. But then, when their AI tools weren’t available, they underperformed.
These findings are also reflected in the work environment. A 2026 survey by IT firm GoTo found that 39% of workers say their reliance on AI has already weakened their skill sets. Same dynamic. Various settings.
This raises a question that every parent, manager, and professional should ponder: What happens to your ability to think and achieve your goal when AI does the thinking for you? goals?
Effort is an important variable for learning
Psychologists call it cognitive overload. This is what happens when you externalize your thinking to save your mental energy. Using a calculator or setting reminders are all forms of cognitive overload that free up mental bandwidth for higher-level thinking. The strategy becomes detrimental when the offloaded task is specifically designed to promote learning, such as using AI to complete homework or wind through a workplace training program.
Wharton professor Ethan Mollick, who has studied AI adoption Across settings, he summed up the main conclusion of the Chinese student study when he said that AI “harms learning if it slows down mental effort.”
The brain creates capacity through struggle. When a problem arises, the mind draws on previous knowledge, makes new connections, and develops its type pattern recognition which does not constitute any other way. This judgment is formed through moments when the answer is not obvious and your brain has to come up with the answer.
Tools that remove effort slow down learning and growth.
This is where the top students suffered the most. They were more able to delegate their work to AI. Nothing in their homework scores indicates a problem. In fact, their homework was of high quality because they instructed the AI at a high level. Paradoxical as it may seem, the students who benefited most from AI were also negatively affected.
Invisible time
What complicates the effect of cognitive loading is its latency. Homework results look better than ever. But the decline in learning did not show up in the students for months, and the full effect took about two years.
The lag creates a feedback loop that makes the AI harder learning problems. When the output is strong, there is no clear signal that the behavior needs to change. By the time the difference becomes apparent—for example, after a poor performance on an exam, a job interview, or any moment that requires you to think on your feet—the learning curve has already imperceptibly increased.
MIT Sloan researchers model a dynamic at work they call a “growth trap”: short-termism labor productivity Achievements remove experience, these achievements are related over time. In other words, workers who use AI will achieve good results initially, but will lose the capabilities that allow for long-term success with AI. Data from GoTo research shows that among Gen Z workers, 46% say that relying on AI is weakening their skills. This is a big challenge, as most Gen Z workers are currently undergoing foundational training and judgment that their careers depend on the future.
What leads to real learning
The Chinese study also found that students who used AI showed minimal learning loss while maintaining strong effort. Using AI as a tool, not a crutch, allowed students to absorb and retain learning.
When it comes to artificial intelligence, the antidote is for humans to work alongside AI, which will eventually augment human capabilities. Treat AI as a contribution to thinking, not a replacement for it.
Do less and it becomes a story about itreconnaissance“It really is and where it came from. As I describe in my latest book, Experimental intelligenceBeing “smart” is a combination of IQ (intelligence), EQ (emotional intelligence) and wisdom from real life experiences. Experiences shape your thinking and abilities in a way that develops judgment, a skill that is needed more than ever in today’s artificial world. Students who worked alongside AI actually gave themselves a stronger learning experience while using it than those who used it to avoid the homework process.
Whether you are a parent, teacher or manager, consider this question for yourself and those you support: Where AI is helping you get results today, do you still think that your ability will increase over time, or will you get better results and quietly depend on the tool that generates them?
