在以人机交互为诉求的人工智能产品的生产中，与识别、判断、创新等人类能力相关的认知劳动成为主导性的劳动形态。在数据标注工作中，认知劳动的劳动控制目标是使标注员的自然认知转化为与计算机程序相匹配的产生式认知，以高效生产出符合电脑需要的标准化数据。其中，管理方通过认知标准化、认知反馈与认知加速，巧妙地整饰着数据标注员的认知行为；后者则通过产量限制、离职等方式对劳动异化进行抵制。从体力劳动到认知劳动，管理控制的对象从劳动者肢体运动方式转变为其大脑认知方法，劳动控制方式则从机械的身体规训走向了灵活的认知整饰。

In the production of artificial intelligence products that appeal to humancomputer interaction, cognitive labor related to human abilities such as recognition, judgment and creation becomes one of the dominant labor forms. In data labeling work, the labor control goal of cognitive labor is to transform the natural cognition of the labeler into generative cognition that matches the computer program to efficiently produce standardized data that meets the needs of the computer. In this regard, management subtly manipulates the cognitive behavior of data labelers through cognitive standardization, cognitive feedback and cognitive acceleration; the latter resists alienated labor by means of output restrictions, quarrels and job hopping. From physical to cognitive labor, the object of management control changes from the way workers move their bodies to the cognitive pattern of their brains, and labor control strategies move from mechanical body discipline to flexible cognitive guide.

国家社会科学基金项目“人工智能产业中的算法劳动与劳动关系研究”（项目编号：20CSH046）。

欢迎阅读《社会学研究》！您是该文第2857位读者！