大学生生成式人工智能应用行为的双维结构与使用现状研究——基于问卷调查的实证分析

Research on the Two-Dimensional Structure and Current Characteristics of College Students’ Generative Artificial Intelligence Application Behavior—An Empirical Analysis Based on Questionnaire Survey

摘要: 本研究基于108份有效问卷，尝试采用描述统计与探索性因子分析，探索大学生生成式人工智能应用行为的维度与特征。研究发现如下：首先，因子分析结果显示大学生生成式人工智能应用行为确实具有双维结构，相关题项可提取为“深度协同与审慎使用”和“替代式依赖使用”两个因子。其次，目前大学生群体的生成式人工智能应用已较为普遍，样本中有75.00%的受访者每天至少使用一次生成式人工智能工具，应用场景以写作、学习辅助或查证信息为主；再次，大学生生成式人工智能应用的大致格局为“深度协同为主、替代依赖为辅”，少数学生存在直接套用、不核验生成内容的风险行为。此外，群体差异分析显示，使用频率越高者深度协同得分越高、替代式依赖反而越低；外部要求与内在兴趣动机较强者的深度协同得分更高，但动机对替代式依赖基本无预测力。基于研究结论，提出以下建议：高校人工智能素养教育不应局限于“能否使用生成式人工智能”的二元对立思维，而应聚焦培育学生“如何合理使用生成式人工智能”的能力，设计分层引导方案。

Abstract: Based on 108 valid questionnaires, this study attempts to explore the dimensions and characteristics of college students’ generative artificial intelligence application behavior, using descriptive statistics and exploratory factor analysis. The findings are as follows: First, the factor analysis results indicate that college students’ generative artificial intelligence application behavior indeed exhibits a two-dimensional structure, with the relevant items extracted into two factors—“Deep Collaboration and Prudent Use” and “Substitutive Dependent Use”. Second, generative artificial intelligence applications have become relatively widespread among college students today, with 75.00% of the respondents in the sample using generative artificial intelligence tools at least once a day, mainly for writing, learning assistance, and information verification. Third, the overall pattern of college students’ generative artificial intelligence application is characterized by “deep collaboration as the dominant mode, substitutive dependence as the secondary one”, while a small number of students display risk behaviors such as directly copying outputs without verifying the generated content. Fourth, the group-difference analysis shows that higher usage frequency is associated with higher deep-collaboration scores and lower substitutive dependence; respondents with stronger external-requirement or intrinsic-interest motivation score higher on deep collaboration, whereas motivation shows no significant predictive effect on substitutive dependence. Based on these conclusions, the following suggestions are proposed: AI literacy education in universities should not be confined to the binary opposition of “whether to use generative artificial intelligence”, but should instead focus on cultivating students’ capacity for “how to use generative artificial intelligence appropriately” and designing tiered guiding strategies.