从数字黑箱到思维熔炉：生成式人工智能环境下算法课程的质量分析与重构策略

From Digital Black Box to Crucible of Thought: Quality Analysis and Restructuring Strategies for Algorithm Courses in the Age of Generative AI

摘要: 生成式人工智能的普及对高等教育构成系统性影响，算法设计与分析课程因与AI训练数据生态深度耦合，面临比其他计算机课程更突出的挑战。本文旨在揭示这一结构性困境的成因、表现与应对路径。本文通过剖析LLM训练语料构成，揭示算法课程与AI之间的“数据同源性”；引入“数据自噬”概念解释知识生产生态失衡风险；基于2023~2024与2024~2025两届学生的课堂追踪对比，分析AI使用行为、知识内化效果与评价体系失效情况。通过追踪发现学生已从“审慎尝试”进入“深度嵌入”阶段，使用率从约45%上升至85%以上，但知识内化显著分化；高水平学生能从AI输出中逆向学习，大量学生陷入“理解债务”；传统编程作业、刷题题库与知识复述类考题在AI环境下区分度明显下降。针对上述问题，本文提出课程重构须将AI从“作弊工具”转化为“教学显微镜”，以“作业即对话、考试即辩证、课堂即共建、AI即陪练”四维策略，使算法课程成为培养计算思维的“思维熔炉”。

Abstract: The rapid growth of generative artificial intelligence has a systemic impact on higher education. Compared with other computer science courses, the Algorithm Design and Analysis course faces particularly acute challenges due to its deep coupling with AI training data ecosystems. This paper aims to uncover the causes, manifestations, and response pathways of this structural crisis. To this end, we deconstruct the composition of LLM training corpora to reveal the “data homology” between algorithm curricula and AI, and introduce the concept of “data autophagy” to explain the erosion of the knowledge production ecology. Through a comparative longitudinal observation of two student cohorts (2023~2024 and 2024~2025 academic years), we analyze shifts in AI usage behavior, knowledge internalization outcomes, and the failure of assessment systems. Our findings show that students have transitioned from a phase of “cautious experimentation” to one of “deep embedding”, with usage rates surging from approximately 45% to over 85%, while knowledge internalization has diverged sharply: high-performing students demonstrate reverse learning from AI outputs, whereas a substantial number of students fall into an accumulating trap of “comprehension debt”. Meanwhile, conventional programming assignments, drill-based problem bank items, and knowledge-recall examination questions have largely lost their discriminative power in the AI context. In response to these challenges, this paper proposes that curriculum restructuring must transform AI from a student “cheating tool” into a “pedagogical microscope”, and adopts a four-dimensional strategy: “assignments as dialogues, examinations as dialectics, classrooms as co-construction, and AI as a practice partner”, to enable the algorithm course to function as a genuine “crucible of thought” for cultivating computational thinking.