Using AI can create short-term results but may stifle long-term learning. #
Posted June 25, 2026 [ Reviewed by Jessica Schrader
](/us/docs/editorial-process)
Key points
- Using AI to “think for you” can erode learning.
- Top performers face the biggest learning losses from AI use.
- Putting in effort along with AI can lead to greater learning.
A new study tracked 26,811 Chinese secondary school students over 30 months. The research measured how generative AI impacted their learning.
*The good news: *Students using AI saw homework scores rise 18% and completion time drop by about 30%. Better scores done faster. These are outcomes any educator should be happy to see.
The bad news: The same students saw their closed-book exam scores fall 20% within six months. Entrance exam scores dropped between 18% and 24%, with the greatest declines surfacing after roughly two years of AI use.
About 80% of student AI users demonstrated the results of outsourcing thinking. They finished assignments unusually fast. They scored well on homework. But then they underperformed when their AI tools weren’t available.
These findings also show up in the work setting. 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. Different settings.
Which raises a question every parent, manager, and professional should consider: When AI does the thinking for you, what happens to your ability to think, and achieve your goals?
Effort Is the Critical Variable for Learning #
Psychologists call it cognitive off. That’s what happens when you outsource your thinking to an external tool to save your mental energy. Using a calculator or setting reminders are all forms of cognitive off that free up mental bandwidth for higher-level thinking. The strategy becomes harmful when the offloaded task is designed specifically to promote learning, like using AI to do homework or breeze through a workplace training program.
Wharton professor Ethan Mollick, who has studied AI adoption across settings, summarized the China student study’s core finding when he said that AI “hurts learning if it undermines mental effort.”
The brain builds capability through struggle. When a problem arises, the mind draws on prior knowledge, makes new connections, and develops the kind of pattern recognition that doesn’t form any other way. That judgment builds through the moments when the answer isn’t obvious and your brain has to earn the answer.
Tools that eliminate effort slowly erode learning and growth.
This is why the highest-achieving students suffered the largest losses. They were most capable at outsourcing their work to AI. Nothing in their homework scores indicated the problem. In fact, their homework was high quality because they instructed the AI at a superior level. As paradoxical as it sounds, the students who were best positioned to benefit from AI were also the most negatively impacted.
The Invisible Timeline #
What makes the impact of cognitive off difficult to catch is its delay. Homework scores appeared better than ever. But the learning decline didn’t surface for months in students, and the full impact took nearly two years to appear.
The lag creates a feedback loop that exacerbates the AI learning problem. When output looks strong, there’s no clear signal that behavior needs to change. By the time the gap becomes apparent—for example, after a poor performance on an exam, a job interview, or any moment requiring thinking on your feet—the learning decline has already been invisibly compounding.
MIT Sloan researchers modeling the same dynamic in the workplace called it the “augmentation trap”: short-term productivity gains erode the expertise those very gains depend on over time. Said another way, workers who use AI get great results initially, but diminish the capabilities that allow for longer-term success with AI. The data from GoTo’s research reveals that among Gen Z workers, 46% say AI reliance is weakening their skills. That’s a big problem given most Gen Z workers are currently building the foundational learning and judgment their careers will depend on for the future.
What Leads to Real Learning #
The China study also found that students who used AI while still making a strong effort showed minimal learning loss. Using AI as a tool, not a crutch, allowed students to internalize and keep the learning.
When it comes to AI eroding learning, the antidote is for people to work alongside AI in ways that build human capability over time. Treat AI as an input to thinking rather than a replacement for it.
Artificial IntelligenceEssential Reads Zoom out, and this becomes a story about what “intelligence” actually is and where it comes from. As I describe in my latest book, Experiential Intelligence, being “smart” involves a combination of IQ (intellect), EQ (
[emotional intelligence](https://www.psychologytoday.com/us/basics/emotional-intelligence)) and the
[wisdom](https://www.psychologytoday.com/us/basics/wisdom)gained from real-life
experiences. Experiences shape your mindsets and abilities in ways that result in developing judgment, a skill needed more than ever in today’s AI-powered world. The students who worked alongside AI essentially gave themselves a more robust learning experience while using it, versus those who leveraged it to sidestep the homework process itself.
Whether you’re a parent, educator, or manager, consider this question for yourself and those you’re supporting: In the places where AI is helping you produce results today, are you still doing the thinking that builds your capacity over time, or are you producing better outputs while quietly becoming more dependent on the tool that generates them?
References
Caosun, M., & Aral, S. (2026). The augmentation trap: AI productivity and the cost of cognitive off. arXiv. https://arxiv.org/abs/2604.03501 GoTo & Workplace Intelligence. (2026). The pulse of work in 2026: Opportunity, risk, and responsibility in an AI-driven workplace.
Kaplan, S. (2023). Experiential Intelligence: Harness the power of experience for personal and business breakthroughs. Matt Holt Books.
Risko E, Gilbert S. (2016). Cognitive OffTrends in Cognitive Sciences, 20, 676-688
Strömberg, D., Lei, V., & Wu, Y. (2026). The generative AI learning penalty: Evidence from Chinese secondary education (CEPR Discussion Paper No. DP21577). Centre for Economic Policy Research.