社会杂志 ›› 2025, Vol. 45 ›› Issue (2): 1-31.

• 专题一：新型劳动关系研究 •    下一篇

智能制造甘愿：人工智能训练的劳动组织形式与控制策略

黄晖()   

  • 出版日期:2025-03-20 发布日期:2025-04-29
  • 作者简介:黄晖  上海交通大学国际与公共事务学院, E-mail: hui.huang@sjtu.edu.cn
  • 基金资助:
    国家社科基金青年项目“数字时代新职业群体的社会融合困境与对策研究”(23CSH027)

The Making of Consent to Produce AI: Labour Organisation Forms and Control Mechanisms in Data Annotating Industry

Hui HUANG()   

  • Online:2025-03-20 Published:2025-04-29
  • About author:HUANG Hui, School of International and Public Affairs, Shanghai Jiao Tong University, E-mail: hui.huang@sjtu.edu.cn
  • Supported by:
    the National Social Science Fund of China, Youth Project(23CSH027)

RichHTML

50

PDF

379

摘要：

日前，生成式大模型的轰然问世引发社会对其“智能”背后的密集型不稳定劳动的关注。基于笔者对三家不同类型的人工智能公司的劳动社会学考察，本文探究了人工智能大模型生产的底层劳动组织形式及其控制策略。研究发现，人工智能训练以项目制为核心，构建了一种以内包、外包和众包为主要组织形式的灵活劳动体制。尽管该劳动体制削弱了劳动者的自主性，加剧了劳动的不稳定性，但资本通过数智游戏、技能清零、职业复魅等控制手段向人工智能训练师施加霸权权力，使之甘愿在不稳定的劳动关系中通过“让人变成机器的工作让机器变得像人”。

关键词: 人工智能, 项目制, 不稳定劳动, 劳动组织, 劳动自主性

Abstract:

The rapid proliferation of generative AI models has sparked critical inquiry into the hidden precarious labour infrastructures that help sustaining their performance. This article draws on ethnographic research conducted in three Chinese AI companies to examine how the production of large-scale models is made possible through intensive, low-paid and precarious data work. It argues that AI production is underpinned by a project-based labour regime structured with insourcing, outsourcing and crowdsourcing as its main organizational forms. The regime has systematically weakened the autonomy of labor, exacerbated the instability of labor, and presented significant characteristics of labor alienation. Rather than overt resistance, workers tend to display consent and acceptance of precarious conditions. In order to conceal the essence of its labor exploitation, capital employs three main strategies of normative control to exert hegemonic power over labor in order to create "willingness" on the part of labor. This study explores how such consent is being actively produced. Gamification mechanisms reframe exploitative work as cognitively stimulating and competitive; task modularisation and fast-changing project cycles lead to cyclical deskilling, curbing worker leverage and occupational mobility; and the symbolic valorisation of AI work fosters a sense of meaning and belonging in otherwise marginal roles. These mechanisms operate as technologies of consent, embedding hegemonic control within the everyday organisation of AI labour. This paper uncovers the paradoxical reality in contemporary AI production: how capital manufactures consent to "make human work like machines so that machines can appear more human". The findings extend classic labour process theory and contribute to a deeper understanding of labour organisation and control mechanisms in the age of artificial intelligence.

Key words: artificial intelligence, projectification, precarious work, labour organisation, work autonomy