打破“灵感神话”：从线性依赖到递归协作的AI辅助写作教学设计与实践

Breaking the “Myth of Inspiration”: AI Assisted Writing Teaching Design and Practice from Linear Dependence to Recursive Collaboration

摘要: 生成式人工智能在高等教育写作中迅速普及，写作效率获得了极大提升，但同时也让教师们产生了教学忧虑：学生是否正从“学习写作”滑向了“依赖AI生成”？当前多数研究聚焦于“写作辅助”，忽视了更根本的“学习辅助”问题。本研究基于Flower & Hayes的写作认知递归理论，整合动态支架理论，以“生成–比较–反思–修正”四轮递归教学单元开展教学实践，力求构建“认知–技术双递归”写作教学模型。研究对700名本科生进行了问卷调查，并通过教学实践收集修改轨迹包和反思日志等学习数据。调查发现，50%的学生将主要写作障碍归于“缺乏灵感”，而85.43%认为AI增加了“创作灵感”——这一“灵感悖论”反映出学生陷入“灵感神话”的认知偏差。教学实践显示，四轮递归教学明显提升了学生的高阶修改行为占比，促进了从“AI帮我写”到“我借AI学”的观念转变。本研究将从“线性依赖”到“递归协作”的转变机制归纳为外化、比较、归责三个关键环节。

Abstract: Generative artificial intelligence has rapidly become popular in higher education writing, greatly improving writing efficiency. However, it has also raised concerns among teachers about whether students are sliding from “learning writing” to “relying on AI generation”? Most current research focuses on “writing assistance” and neglects the more fundamental issue of “learning assistance”. This study is based on Flower & Hayes’ writing cognitive recursion theory, integrating dynamic scaffolding theory, and conducting teaching practice with a four round recursive teaching unit of “generation comparison reflection correction”, aiming to construct a “cognitive technical dual recursion” writing teaching model. A questionnaire survey was conducted on 700 undergraduate students, and learning data such as modification trajectory packages and reflection logs were collected through teaching practice. A survey found that 50% of students attribute their main writing barrier to “lack of inspiration”, while 85.43% believe that AI has increased “creative inspiration” - this “inspiration paradox” reflects students’ cognitive bias into the “myth of inspiration”. Teaching practice has shown that the four round recursive teaching significantly improves the proportion of students’ higher-order modification behaviors, promoting a shift in mindset from “AI helps me write” to “I borrow AI to learn”. This study summarizes the transition mechanism from “linear dependence” to “recursive collaboration” into three key steps: externalization, comparison, and attribution.