生成式人工智能何以融入质性研究——以“生成田野”为起点的功能评价与伦理审思  

How Generative Artificial Intelligence Can Be Integrated into Qualitative Research - Functional Evaluation and Ethical Reflection Starting with“Generated Fieldwork”

作　　者：马颂歌[1,2] 左靖文 Ma Songge;Zuo Jingwen(Schoo of Education,Shanghai Normal University,Shanghai 200234)

机构地区：[1]上海师范大学教育学院,上海200234 [2]上海师范大学终身教育政策研究中心,上海200234

出　　处：《远程教育杂志》2024年第6期83-91,共9页Journal of Distance Education

基　　金：全国教育科学“十四五”规划2023年度国家青年项目“社会技术理论视域下老年数字能力形成机制及教育支持路径研究”(课题批准号:CKA230333)的阶段性成果。

摘　　要：生成式人工智能技术背景下,研究者已经创造出虚拟了物理环境、社会环境和人际关系的“生成田野”。个案分析发现,在开放式深度访谈中,人工智能受访者需要全面的人物设定,其生成的访谈文本具有“对应倾向”“常识回应”和“关系模糊”特性,说明其信息供给能力强、关系供给能力弱。在阅读、摘录、分类等基础编码环节,人工智能编码的效率优势明显,但有“总结倾向”和“分析倾向”。在叙事、解释、分析等解析编码环节,人工智能表现出用学界常识批判文化持有者内部眼界的倾向,对本土概念解释失准,未能发现常识性悖论。投喂常识性悖论后,人工智能即可生成全面、系统、有新意的理论分析。基于人工智能在质性研究各阶段之“能为”,研究者应思考是否向人工智能让渡自身研究权利的“允为”问题,学界在“应为”方面可首先开放利用人工智能进行模拟的权限,允许研究者“真实”与“虚拟”双线并行,借虚拟访谈训练技术、适应情境,借虚拟分析提供线索、生发灵感。With the advancement of generative artificial intelligence(AI),researchers have developed a“generative field”that digitally simulates physical,social,and interpersonal environments.Case studies reveal that AI-generated interviewees require comprehensive character profiles in open-ended,in-depth interviews.The interview responses generated by AI exhibit characteristics such as“correspondence bias,”“commonsense response bias,”and“relational ambiguity,”indicating AI’s strong capability for information provision but limited ability to foster meaningful relational dynamics.In basic coding tasks such as reading,excerpting,and categorizing,AI demonstrates significant efficiency advantages,though it tends to exhibit“summary bias”and“analytic bias.”During more interpretive stages of coding,including narration,explanation,and analysis,AI often relies on academic norms to critique the internal perspectives of cultural knowledge holders,leading to misinterpretations of indigenous concepts and a failure to identify paradoxes in commonsense reasoning.However,when provided with such paradoxes,AI can generate comprehensive,systematic,and novel theoretical analyses.Given AI’s varied capabilities at different stages of qualitative research,scholars must carefully consider whether to delegate certain research responsibilities to AI,addressing ethical questions of what should be permitted.It is recommended that academics begin by allowing AI to support simulations,enabling researchers to conduct“real”and“virtual”parallel studies,utilizing virtual interviews for skill development and contextual adaptation,while also leveraging virtual analyses to inspire fresh ideas.

关 键 词：质性研究方法 生成式人工智能 人工智能编码 人工智能访谈 生成田野 

分 类 号：G40-032[文化科学—教育学原理]