Summary of Key Points
Generative AI has become increasingly integrated into the learning process of primary and secondary school students, with Norway introducing the world's strictest policies to address potential risks (banning its use for primary school students and restricting it for middle school students). A research team from the University of Hong Kong has discovered a phenomenon known as "learning punishment" due to AI: students using AI to complete assignments see short-term improvements in efficiency (grades increase by 18%, and completion time is reduced by 30%), but their cognitive abilities and test scores decline over the long term (monthly exams drop by 20% after 6 months, with the highest drops observed in middle school entrance exams and college entrance exams). The effects are particularly pronounced among the top 1/3 of students, boys, and in subjects involving social sciences. Professor Wu Yanhui emphasizes that AI is not inherently harmful; the key lies in distinguishing between tasks that can be delegated to AI and those that require human skills. He calls for a redefinition of learning and a reform of the education evaluation system to cultivate abilities that AI cannot replace.
1. Does "learning punishment" really exist? Data provides the answer
The University of Hong Kong team tracked the academic performance of over 26,000 middle and high school students in nine schools in a central county over a period of 30 months, comparing students who used AI with those who did not:
- Short-term benefits: Students using AI saw an average 18% increase in assignment scores and a 30% reduction in completion time.
- Long-term consequences: Monthly exam scores dropped by 20% after 6 months, and by up to 24% in middle school entrance exams and 18% in college entrance exams after two years.
- Highly affected groups: Students in the top 1/3 of the class (who lose the ability to think critically about difficult problems), boys (who are more likely to rely on AI for homework), and students in social science subjects.
- Differences in usage: 81% of students used AI merely to copy answers, resulting in significant score losses, while a minority used it as a guide to enhance their independent thinking, with smaller impacts.
(Note: The study is a retrospective analysis, not real-time observation, but its conclusions are still cautionary.)
2. Why are top students more affected by AI?
Professor Wu explains that exams differentiate students based on their ability to solve complex, thought-provoking problems. In the past, top students excelled by tackling these challenges independently, which became their core strength. With AI, these thinking processes are outsourced, leading to a decline in their abilities and, consequently, in their scores. For example, a difficult math problem that used to take a top student an hour to solve now gets an answer from AI in just one minute, but without understanding the solution method, the student may still struggle when faced with similar problems in exams.
3. What does "learning" really mean in the AI era?
- Completing assignments does not equal learning: Using AI to quickly complete assignments provides no real knowledge or skill acquisition; these tasks completed by AI are meaningless in the real world.
- The rationale behind Norway's policy: Primary school students have not yet developed fundamental cognitive skills such as reading, calculation, and logical thinking, which are essential human capabilities. Relying on AI too early can be detrimental.
- The paradox in China: Educational goals are still focused on the college entrance exam, and schools prioritize using AI to boost scores. However, a balance between short-term gains and long-term skill development is needed. For instance, a school in Shenzhen requires students to use AI for previewing lessons but then write out the complete solution steps, which, although more time-consuming for teachers, encourages students to think critically.
4. How should the education evaluation system be reformed?
The current system, which focuses on scores, is at odds with the use of AI. Possible reforms include:
- Increasing in-class quizzes: Conducting short, unassisted quizzes daily to ensure students have truly mastered the material.
- Adjusting evaluation criteria: Moving from focusing on the accuracy of answers to assessing creativity, critical thinking, and problem-solving abilities.
- Reforming the college entrance exam: If AI can replace rote learning, future exams should focus on skills that AI cannot provide, such as humanistic literacy, teamwork, and innovative thinking.
5. What abilities should humans retain that AI cannot replace?
Professor Wu cites a new project from the University of Hong Kong to highlight the future core competencies:
- Understanding AI: Knowing how AI works (mathematics, programming, machine learning) to avoid being controlled by it.
- Understanding society: Comprehending social rules through economics and other fields to see how AI affects humans.
- Understanding human nature: Cultivating abilities beyond intelligence through humanities and arts (e.g., the Book of Songs, Homer’s epics) to appreciate beauty, pursue interests, and develop intrinsic motivation.
- Critical abilities: The ability to ask meaningful questions (AI needs humans to generate good questions), the judgment to decide when not to use AI, and social skills (which AI cannot replace).
The ideal scenario is for AI to handle routine, repetitive tasks, while humans focus on creativity, critical thinking, and exploration. However, this requires a concurrent reform of the education system. Otherwise, the use of AI may lead to a "score-boosting arms race" that undermines the purpose of education.
The impact of AI on education is both a challenge and an opportunity. It forces us to reconsider the purpose of education: to cultivate individuals who are not just capable of solving problems but also warm, thoughtful, and creative.