研究显示：AI 助力家庭作业得分提高了 18%，但随后的考试成绩却下降了 20%

研究显示：AI 助力家庭作业得分提高了 18%，但随后的考试成绩却下降了 20%
AI boosted homework scores, then exam scores dropped: Study

由斯德哥尔摩大学和香港大学研究人员进行的一项针对 27,000 名中国学生的研究显示，人工智能辅助作业的表现与实际学习成果之间存在显著差异。在六个月的时间里，使用人工智能工具的学生作业成绩提高了 18%。然而，当在没有人工智能辅助的闭卷环境下进行测试时，这些学生的得分比未使用该技术的同龄人低 20%。这表明，虽然人工智能有助于完成即时任务，但可能会阻碍核心认知技能的发展和长期知识的保持。这些研究结果与教育界更广泛的担忧一致，即过度依赖聊天机器人会导致作业“程式化”，且缺乏对学科内容的深入参与。宾夕法尼亚大学 2024 年的一项小型研究也得出了类似的结论，指出人工智能在练习环节带来的优势未能转化为考试成功。尽管人工智能可以作为有益的教育补充，但布鲁金斯学会的专家警告称，用自动化工具取代批判性思维和社会互动，可能会阻碍学术发展。研究人员的发现强调，在课堂中整合人工智能时，需要采取平衡且审慎的方法。

一项最新研究表明，虽然人工智能工具可以将家庭作业成绩提高 18%，但可能会导致考试成绩下降 20%。该报告在 Hacker News 上引发了关于人工智能在教育和职业发展中作用的热烈讨论。批评者认为，过度依赖人工智能会阻碍学生建立基础知识、培养批判性思维以及在没有数字辅助的情况下独立工作的能力，这可能会造就一代“文盲”专业人士。相反，一些人认为传统的作业方式已经过时，社会应该优先考虑实际应用能力的掌握，而非死记硬背。这场讨论凸显了两种理念之间的矛盾：一种认为学习过程中的“挣扎”对于大脑发育和职业自主性至关重要；另一种则认为人工智能是必然的趋势，它使传统的学术劳动变得多余。最终，参与者发出警告称，学习过程的自动化可能会导致人类直觉的丧失，并使人们过度依赖算法来定义“最佳实践”，这引发了人们对人类能力长期受损以及社会对人工智能依赖性的担忧。

原文

8,897

A new study tracking 27,000 students in China has found that pupils who used artificial intelligence tools saw higher homework scores over time, but performed worse than their peers on exams taken without AI assistance, according to research covered by The Economist on August 18.

The Study

The research was conducted by David Stromberg of Stockholm University along with Victor Lei and Wu Yanhui of the University of Hong Kong. The study followed 27,000 pupils aged 12 to 18 in China, where adoption of AI tools among students has grown quickly. Around 80% of the students surveyed reported using AI models such as Doubao and DeepSeek, while the remaining 20%, who did not use such tools, formed a control group.

According to figures shared by The Economist, students who used AI saw their average homework scores rise by 18% across all subjects over a six-month period. However, when the same students were tested under exam conditions without access to AI tools, they scored 20% below classmates who had not used AI during the study period.

Context on AI Adoption Among Students

The study cited broader data on how widespread AI use has become among students. A survey conducted last year by ed-tech firm Chegg found that 80% of undergraduate students in wealthy countries reported using AI in their studies. More recent polling put the figure at 94% among students in Britain and 93% in Germany.

The Economist noted that teachers have reported grading formulaic, similar-sounding essays they suspect were generated by AI chatbots such as ChatGPT, but said that, prior to this study, robust evidence on AI’s actual effects on learning outcomes had been limited.

Related Research

A separate study conducted in 2024 at the University of Pennsylvania examined a similar dynamic on a smaller scale. Students attending a math lesson practiced problems using either traditional study methods, such as notes and textbooks, or AI tools including ChatGPT and an AI tutoring program. According to a summary of the research, students using AI performed better during short-term practice sessions, but the advantage did not carry over to a subsequent closed-book test.

Reactions

The Economist’s summary of the findings, shared on the social platform X, drew significant engagement, with some commenters attributing the exam score gap to students copying AI-generated answers without engaging deeply with the material. The study was also discussed on forums including Hacker News, where some users questioned aspects of the study’s design while others noted it aligned with existing concerns among educators.

A tip sheet published by the Brookings Institution earlier this year said AI can support learning when used intentionally and designed well, but cautioned that overreliance on the technology to replace thinking, social interaction, or creativity could prevent students from developing cognitive and social skills.

The Stockholm University and University of Hong Kong researchers’ full study had not been independently verified by other institutions at the time of publication.