【摘要】 在生成式人工智能参与社会科学知识生产过程中，提示词工程通常被认为是一种提高大模型回应质量的关键路径，但提示词工程存在诸多缺陷，它在形式上承担着引导大模型输出的任务，实质上却被简化为指令式、片段化、非结构化的语句，难以真正引导深度知识的生长。“问力”是研究者的一种复合型认知能力系统，具体包括问题识别能力、语义组织能力、追问递进能力、交互适配能力和知识反思能力。“问力”与提示词工程存在本质区别，大模型的真正潜力，不应仅由提示词的操作者调动，而应由具备高阶“问力”的研究者激活。“问力”深刻影响知识生产的深度和广度。“问力”引导的人机互动知识生产标志着社会科学研究范式的转型与跃迁，它催生了“人—问—机”三元协同的知识生成图式。更多还原

【Abstract】 In the process where generative AI participates in the social sciences knowledge production, prompt engineering is commonly regarded as a key approach to improving the response quality of large language models. However, prompt engineering has numerous limitations. While formally tasked with guiding large models’ outputs, it is essentially reduced to directive, fragmented, and unstructured statements, failing to genuinely facilitate the generation of in-depth knowledge. The “Questioning Ability” refers to a composite cognitive capability system developed by researchers, encompassing problem identification, semantic organization, progressive questioning, interactive adaptation, and knowledge reflection. This system fundamentally differs from prompt engineering. The true potential of large models should not be activated solely by prompt operators but rather by researchers equipped with advanced “Questioning Ability”. The Questioning Ability profoundly influences the depth and breadth of knowledge production. Human-machine interactive knowledge production guided by the Questioning Ability marks a transformation in social science research, giving rise to a tripartite collaborative knowledge generation framework of “human-question-machine.”更多还原