Human-Machine Collaboration and the Transformation of the Digital Capitalist Labor Process under Large-Model Production

Entering the era of large-model production, generative artificial intelligence is driving the accelerated iteration of intelligent machines with lightning speed, exhibiting subject-characteristics different from previous artificial intelligence and forming a completely new form of human-machine interaction. From the perspective of the production process of digital capital, intelligent machines and knowledge workers each leverage their strengths and collaborate effectively, fundamentally reshaping the nature of work and production processes, and continuously enacting a new scene of “human-machine co-dance.” On the surface, human-machine collaboration under large-model production, through intelligent enhancement and human-machine co-creation, liberates people from monotonous, repetitive, and tedious labor, allowing them to focus on more creative work or tasks. Intelligent machines have suddenly become humanity’s capable assistants, work partners, mentors, and even emotional companions, and the capitalist contradiction of “the iron man against the man of flesh and blood” [n1] appears to have been “repaired” once and for all. So much so that one research report bluntly declares: “For the first time in human history, we have welcomed a ‘humanized’ new generation of technology.”

进入大模型生产时代，生成式人工智能以迅雷不及掩耳之势推动智能机器加速迭代，表现出与以往的人工智能不同的主体性特征，形成了全新的人机交互形态。从数字资本的生产过程来看，智能机器和知识工人各施所长、有效协作，根本性地重塑了工作性质和生产流程，不断演绎着“人机共舞”的新景象。从表面上看，大模型生产下的人机协作通过智能增强和人机共创将人从单调、重复且枯燥的劳动中解放出来，使其专注于更具创造性的工作或任务。智能机器一跃成为人类的得力助手、工作搭档、良师益友甚至情感陪伴，资本主义“铁人反对有血有肉的人”的人机矛盾看似一劳永逸地得以“修复”了。以至于有研究报告直言不讳地宣称：“人类历史上首次迎来了‘人性化’的新一代技术。”

Labor process theory holds that the ideal capitalism strives to realize is dead labor dominating living labor. [n2] Based on this view, the continuously developing study of the labor process aims to “expose the obscured labor process and production scene to the researcher’s field of vision, to return to the ‘hidden places of production’ in order to reveal the relations of oppression and domination hidden within capitalist production.” [n3] Against the backdrop of digital capital’s vigorous construction of a “new world of human-machine collaboration,” a theoretical question urgently awaiting an answer is: Does the promotion of “human-machine reconciliation” and the move toward “co-intelligence” in digital capitalist large-model production signal that workers trapped in the “algorithmic cage” are escaping the fate of digital enslavement, and, through “a Darwinian co-evolution,” [n4] transforming into “cyborgs,” thereby accelerating the rush toward an ideal society where “all beings are unbound”? Undoubtedly, analyzing the new characteristics of the current human-machine relationship and revealing the new means by which digital capital disciplines and absorbs labor are urgent requirements for deepening the understanding of the laws governing the digital capitalist labor process in the era of large-model production.

劳动过程理论认为，资本主义力求实现的理想是死劳动支配活劳动。(1)基于这一观点，不断发展的劳动过程理论研究旨在将“遮蔽的劳动过程和生产场景暴露在研究者的视野之中，以重返‘生产的隐秘之处’来力图揭示隐藏在资本主义生产中的压迫和宰制关系”(2)。在数字资本倾力打造“人机协作新世界”的背景下，一个亟待回答的理论追问是：数字资本主义大模型生产促进“人机和解”、走向“协同智能”(co-intelligence),是否昭示着禁锢于“算法牢笼”的劳动者正在摆脱数字奴役的宿命，并经由“一场达尔文式的协同进化”(3)而化身“赛博格”,从而加速奔向“众生无束”的理想社会?毋庸置疑，剖析当前人机关系呈现的新特点，揭示数字资本规训和吸纳劳动的新手段，是深化对大模型生产时代数字资本主义劳动过程规律性认识的迫切要求。

This article argues that, entering the era of large-model production, the human-machine collaboration occurring in digital factories is essentially the latest practice of digital capital’s “accelerated mechanization” aimed at increasing labor productivity. Intelligent machines are reshaping the workflows of both material and spiritual production under digital capitalism, objectively enabling more and more workers to withdraw from daily mental exertion and engage in more challenging but more “humanized” work or tasks. But dialectically speaking, human-machine collaboration under the control of digital capital does not necessarily bring workers the digital dividends of increased free time, enhanced work meaning, and higher remuneration. In the long run, the large-scale application of intelligent machines will accelerate the devaluation of knowledge workers’ acquired skills, thereby eroding the practical foundation for the generation of human complex cognitive abilities. Moreover, the digital convenience of “push the button” generation can easily lead to an over-reliance on artificial intelligence for knowledge production, thereby eliminating human creativity and triggering the “deskilling” of the digital capitalist labor process.

本文认为，进入大模型生产时代，发生于数字工厂的人机协作本质上是数字资本旨在提高劳动生产率而“加速机械化”的最新实践。智能机器正在重塑数字资本主义物质生产和精神生产的工作流程，客观上使越来越多的劳动者得以从日常的心力操劳中抽身，去从事更有挑战性但更具“人性化”的工作或任务。但辩证地看，数字资本操控下的人机协作并不必然会给人类带来自由时间增加、工作意义提升和薪酬待遇提高的数字红利。从长远来看，智能机器的规模化应用将促使知识工人习得技能的加速贬值，进而侵蚀人类复杂认知能力生成的实践基础。而且，“一键式生成”(push the button)的数字便捷极易导致知识生产过度依赖人工智能，从而消弭人的创造力，引发数字资本主义劳动过程的“去技能化”。

I. Human-Machine Collaboration: A New Trend in the Development of Capitalist Human-Machine Relations

一、人机协作：资本主义人机关系发展的新趋势

Entering the digital capitalist stage, with the increasing anthropomorphization and agency of intelligent machines, “human-machine adaptation” has opened a new phase in the development of capitalist human-machine relations. In this phase, the flexibility and agility of commodity production processes have continuously improved. After the implementation of large-model production, intelligent machines, empowered by generative artificial intelligence technology, have acquired increasingly diverse and powerful “labor capacities.” Humans and intelligent machines interact frequently through platform systems, and human-machine collaboration has become a new trend in the development of digital capitalist human-machine relations.

进入数字资本主义阶段，随着智能机器的拟人化和能动性不断增强，“人机适应”开启了资本主义人机关系发展的新阶段，在这一阶段，商品生产流程柔性化、敏捷化的水平不断提高。在大模型生产落地后，生成式人工智能技术加持下的智能机器获得了日益多样化、日益强大的“劳动能力”,人与智能机器通过平台系统频繁交互，人机协作成为数字资本主义人机关系发展的新趋势。

(I) The Logic of Emergence of Human-Machine Collaboration under Large-Model Production

(一)大模型生产下人机协作的出场理路

On the one hand, under capitalist conditions, the application of machines extends the functions of human organs, enhances the capacity and efficiency of human practical activities, and promotes the development of productive forces and the increase of social wealth. On the other hand, “the capitalist system has promoted the isomorphism of machine technology and social power, deepened the alienation of workers, and caused the separation and opposition of human-machine relations.” [n5] Entering the era of large-model production, driven by the platform revolution, the explosive growth of data, the exponential supply of computing power, and the increasing refinement of algorithms have jointly promoted the “intelligence emergence” and performance generalization of intelligent machines, presenting a refreshing interactive subjectivity. [n6] Capitalist human-machine relations have begun to transform from a mechanical model of one-way adaptation to an organic model of mutual adaptation. With the empowerment of generative artificial intelligence technology, intelligent machines “are no longer passive and inert tools, but a vibrant ‘partner’ ready to interact with people at any time, actively responding to human demands, providing help, and even offering proactive services.” [n7] As the large-model production system matures, intelligent machines “learn,” “think,” “decide,” and “act” in ways increasingly similar to humans. From the trajectory of machine capability development, intelligent machines, long since no longer “huge machine monsters,” have begun to exhibit “quasi-traits” different from those of humans.

一方面，资本主义条件下机器应用延展了人类器官的功能，提升了人类实践活动的能力和效率，推动了生产力的发展和社会财富的增加；另一方面，“资本主义体制推进了机器技术与社会权力的同构化，加深了工人的异化，造成了人机关系的分离和对立”(4)。进入大模型生产时代，在平台革命的推动下，数据的爆炸式增长、算力的指数级供给以及算法的日益精进，合力推动了智能机器的“智能涌现”和性能泛化，呈现出令人耳目一新的交互主体性。(5)资本主义人机关系开始从单向适应的机械模式向相互适应的有机模式转化。在生成式人工智能技术的加持下，智能机器“不再是被动消极的工具，而是一个充满活力、随时准备与人交互、积极应对人的要求并向人提供帮助甚至主动服务的‘伙伴’”(6)。随着大模型生产体系日臻完善，智能机器以更加类似人的方式“学习”“思考”“决策”和“行事”。从机器能力的发展轨迹来看，早已不是“庞大的机器怪物”的智能机器开始展现出不同于人类的“类特质”。

Broadly speaking, intelligent machines excel at “executing repetitive tasks, analyzing massive datasets, and handling routine cases,” while humans have an advantage in “processing ambiguous information, judging tricky situations, and dealing with customer dissatisfaction” and other humanized soft skills. Thus, on the one hand, in production areas with abundant data supply and predictable scenarios, a new round of “machines replacing humans” is unstoppable. At the same time, in workplaces full of uncertainty, humans’ complex cognitive abilities remain irreplaceable, and the breadth and depth of the capitalist human-machine division of labor have further expanded in the era of large-model production. On the other hand, within the value chain of knowledge commodity production, there are still many process links that neither humans nor machines can handle alone, and which cannot be completed through a clear human-machine division of labor. The root causes are, first, a lack of data in work scenarios and the difficulty of encoding situational knowledge, leading to insufficient data training for intelligent machines and poor “labor performance”; second, the limited information processing capacity of workers, whose intuition, experience, and acquired skills are insufficient to achieve ideal task performance.

大体上看，智能机器善于“执行重复任务，分析庞大的数据集，处理常规案例”,而人类则在“处理模糊信息，判断棘手情况，应对客户不满”等人性化软技能方面占优势。如此一来，一方面，在数据供给充沛且情境可预测的生产领域，新一轮“机器换人”势不可当。同时，在充满不确定性的工作场所，人类复杂的认知能力依旧不可替代，资本主义人机分工的广度和深度在大模型生产时代得以进一步扩张。另一方面，在知识商品生产的价值链条中还存在相当多的工序环节，人和机器都难以独自胜任，无法通过明确的人机分工来完成。究其根源，一是工作场景数据匮乏，情景知识难以编码，从而导致智能机器数据训练不足，“劳动表现”不佳；二是劳动者生产信息处理能力有限，仅凭直觉、经验和习得技能不足以获得理想的任务绩效。

Consequently, digital capital takes a different path, fully leveraging the advantages of adaptive processes brought by large-model production to create human-machine collectives, [n8] blending the nearly infinite computing power of intelligent machines with human insight and comprehensive thinking ability to form “hybrid intelligence,” [n9] fundamentally reshaping the workflow of knowledge commodity production, and thereby achieving rapid development of productive forces through human-machine synergy. Therefore, a development trend of the digital capitalist labor process is that digital capital continuously opens up the “missing middle ground” requiring human-machine collaboration, accelerating the abandonment of programmed assembly lines and shifting toward collaborative work between humans and advanced artificial intelligence systems. [n10] In fields such as manufacturing, marketing, medical diagnosis, and financial services, many traditional production processes are being challenged by entirely new work models of human-machine collaboration.

于是，数字资本便独辟蹊径，充分利用大模型生产带来的自适应流程优势，打造人机集合体(7),糅合智能机器近乎无限的运算能力和人类的洞察力、综合思维能力来形成“混合智能”(hybrid intelligence)(8),从根本上重塑知识商品生产的工作流程，进而通过人机协同发力来实现生产力的高速发展。因此，数字资本主义劳动过程的一个发展趋势就是数字资本不断开辟需要人机协作的“缺失的中间地带”,加速摒弃程序化装配生产线，转向人与先进的人工智能系统的协同工作。(9)在生产制造、市场营销、医疗诊断、金融服务等领域，许多传统的生产流程正在受到人机协作的全新工作模式的挑战。

(II) Practical Forms of Human-Machine Collaboration under Large-Model Production

(二)大模型生产下人机协作的实践样态

Entering the era of large-model production, with the empowerment of increasingly rich multimodal technologies, the concrete forms of intelligent machines also vary. From the perspective of the application and development trends of intelligent machines under digital capitalism, functionally diverse intelligent assistants are gradually becoming the work tools of knowledge workers; digital employees are realizing the concretization of knowledge bases, promoting the intelligent evolution of enterprise organizations; and general-purpose intelligent agents “act according to intention,” capable of enacting the ideal scenario of human-machine collaboration (table omitted).

进入大模型生产时代，在日趋丰富的多模态技术的加持下，智能机器的具化形态也各有千秋。从数字资本主义的智能机器应用及发展趋势来看，功能各异的智能助手逐渐成为知识工人的工作神器；数字员工正在实现知识库具象化，助推企业组织智能化演进；通用智能体“按意图行事”,它可以演绎人机协作的理想图景(表略)。

First, intelligent assistants and human–machine collaboration. With the continuous advancement of the vertical deployment of large models, the functions of intelligent assistants launched in various fields range from automatically generating meeting minutes, assisting in document processing and intelligent search, to data augmentation, code completion and inspiration stimulation. This makes generative-AI-driven intelligent machines the “co-pilot” in work scenarios such as office work, marketing, design and healthcare, significantly extending individual labour capacity. In this sense, “just as computers help humans improve production efficiency and mobile phones help humans communicate, artificial intelligence helps humans transcend physical limitations, enabling them to see farther, hear more clearly, remember more firmly and understand more thoroughly.”(10)

其一，智能助手与人机协作。随着大模型垂直化部署的持续推进，各领域推出的智能助手的功能从自动生成会议纪要、辅助处理文档和智能搜索，到数据增强、代码补全和灵感激发等，这让生成式人工智能驱动的智能机器成为办公、营销、设计、医疗等工作场景的“智能副驾驶”,个体劳动能力得到显著延展。从这个意义上讲，“正如计算机帮助人类提升生产效率、手机帮助人类沟通一样，人工智能有助于人类突破身体局限，让自己看得更远、听得更清、记得更牢、理解得更透彻”(10)。

For example, in intelligent healthcare scenarios, an intelligent assistant can “listen” to a patient’s consultation process and immediately generate medical records, analyse the patient’s genomic data and lifestyle habits, thereby formulating more precise treatment plans, enabling doctors to escape “document fatigue” and focus on communicating with patients, thus improving treatment outcomes. In legal affairs scenarios, intelligent assistants can conduct legal research and case retrieval, automatically generate legal documents, complete contract review and risk assessment, and provide real-time online consultation. In digital marketing scenarios, intelligent assistants can automatically generate marketing copy and video scripts, identify business opportunities in real time, and, with natural-language question-and-answer interaction and powerful semantic generation capabilities, transcend traditional mechanised question-and-answer matching, opening up entirely new paths for the “intelligent transformation” of customer service. In the field of human resource management, intelligent assistants can provide customised learning materials and career development plans based on employees’ characteristics, thereby better tapping into employees’ growth potential.

譬如，在智能医疗场景中，智能助手可以“倾听”患者就诊过程并立即生成医疗记录，分析患者的基因组学数据和生活习惯，从而制定更加精准的治疗方案，使医生能够走出“文档疲劳”,专注于与患者沟通，进而提升诊疗效果。在法律事务场景中，智能助手可以进行法律研究和案例检索，自动生成法律文档，完成合同审查和风险评估，并实时提供在线咨询。在数字营销场景中，智能助手可以自动生成营销文案、视频脚本，实时洞察商机，并以自然语言式问答交互和强大语义生成能力跨越传统的机械化问答匹配，为实现客户服务的“智变”开拓全新的道路。在人力资源管理领域，智能助手可以根据员工的特点提供定制化的学习资料和职业发展规划，从而更好地发掘员工的成长潜力。

At present, various intelligent assistants exhibit autonomy and interactivity that traditional production tools cannot match. Although in the market narrative of digital capital, intelligent assistants still appear as “auxiliary tools,” they possess text-processing, logical-computation, code-editing and other capabilities far exceeding expectations, and are increasingly revealing the production trait of “appearing in the form of productive forces themselves,” providing the material basis for opening up human–machine collaboration in digital capitalism. In practice, intelligent assistants cannot yet independently automate specified processes in knowledge-commodity production; instead, they assist in collecting information in a given field, acquiring expert knowledge and providing decision support in the manner of “intelligence augmentation,” helping labourers complete various digital tasks more efficiently. Because the human–machine collaboration in such scenarios is a loose tool-type cooperative relationship, it has not achieved a “two-way advancement” of human and machine capability evolution, and thus is only a preliminary form of capitalist human–machine collaboration in the era of large-model production.

当前，各种智能助手展现出传统的生产工具不可企及的自动性和交互性。尽管在数字资本的市场叙事中，智能助手依然以“辅助工具”的面貌示人，但其具备了远超预期的文本处理、逻辑计算、代码编辑等能力，开始日益显露出“以生产力本身的形式出现”的生产特质，为开启数字资本主义人机协作提供了物质基础。在实践中，智能助手尚不能单独实现知识商品生产指定工序的自动化，而是以“智能增强”(intelligence augmentation)的方式协助收集某一领域的信息，获取专家知识和提供决策支持，助力劳动者更加高效地完成各种数字化工作。由于此类场景中的人机协作是松散的工具型合作关系，没有实现人与机器能力演进的“双向奔赴”,因而只是大模型生产时代资本主义人机协作的初级形态。

Second, digital employees and human–machine collaboration. In the intelligent production of digital capitalism, digital capital competes to develop sector-specific large models, driving the intelligent transformation of enterprise organisation. To further tap the potential of human–machine collaboration, various enterprises undergoing intelligent transformation begin to use proprietary data to train vertical large models, generating customised digital employees, with the aim of concretising and capitalising the “craft knowledge” of commodity production. Digital employees capable of understanding business logic join the labour force one after another, forming strong alliances with labourers according to the needs of specific work scenarios, creating new human–machine team collaboration. Thus, the subject of the division of labour expands from the division among single human subjects to the division and collaboration of dual human–machine subjects.(11)

其二，数字员工与人机协作。在数字资本主义智能化生产中，数字资本竞相开发部门专有大模型，推动企业组织智能化变革。为了进一步挖掘人机协作的潜力，投身智能化转型的各类企业开始利用专有数据训练垂直大模型，生成定制化的数字员工，以期实现商品生产“工艺知识”的具象化、资本化。能够理解业务逻辑的数字员工纷纷加入劳动力大军，根据特定工作场景的需要与劳动者强强联合，形成新型人机团队协同工作。于是，劳动分工主体由单一主体人的分工拓展至人机双元主体分工协作。(11)

In practice, in the field of business decision-making, digital employees leverage their powerful data-computation capabilities to efficiently collect and mine important information such as market trends and company performance, summarising and extracting patterns or regularities not previously discovered, forming “computational knowledge” and business intelligence in real time; humans, on the basis of algorithm recommendations from digital employees, use their own strategic intuition and interest-balancing skills to promote dynamic optimisation of decisions in an environment full of complexity, uncertainty and interest bargaining. In the field of risk control, digital employees monitor business processes in real time, automatically flag abnormal behaviour, and use professional knowledge for algorithmic reasoning to complete automated causal judgment and risk-source identification; humans only need to verify the algorithmic decision results and take action to eliminate risks. In the field of knowledge production, repetitive tasks such as writing, translation, tabulation, charting and code verification are all handed over to digital employees for automatic execution; labourers can break free from the constraints of “technique” and “efficiency,” focusing on more imaginative and challenging digital labour such as creative problem-solving. In other words, digital employees complete the first draft of work based on human prompts, while labourers set goals, modify and adjust AI-generated content (AIGC) to ensure the compliance and usability of labour outcomes.

从实践上看，在经营决策领域，数字员工发挥自身强大的数据运算能力，高效收集和挖掘市场趋势和公司绩效等重要信息，总结提炼以往没有被发现的模式或规律，实时形成“计算知识”和商业智能；人则在数字员工算法推荐的基础上，利用自身的战略直觉和利益平衡术，在充满复杂性、不确定性和利益博弈的经营环境中促进决策动态优化。在风险控制领域，数字员工实时监控业务流程，自动提示异常行为，并利用专业知识进行算法推理，以完成自动化因果判断和风险源识别；人只需验核算法决策结果，并采取行动消除风险。在知识生产领域，撰文、翻译、制表、作图、代码核查等重复性较高的目标任务，都被交给数字员工来自动执行；劳动者可以突破“技法”和“效能”的限制，专事于创意问题求解等更具想象力和挑战性的数字劳动。换言之，数字员工根据人工提示词完成工作初稿，劳动者则进行目标设定，修改调整人工智能生成内容(AIGC)以确保劳动成果的合规可用。

Compared with the intelligent assistants described above, the “labour performance” of digital employees in many aspects of knowledge production is already on a par with that of humans. Precisely for this reason, in the intelligent transformation of enterprise organisation driven by digital capital, intelligent machines begin to appear as “work partners” jointly solving tasks, objectively exerting a subversive impact on the capitalist labour process. This impact mainly includes three aspects. First, from the perspective of labour status, in human–machine team collaboration, although digital employees cannot complete tasks independently, they bear a workload comparable to that of humans, and are naturally no longer “supporting roles” in the production process. Second, from the perspective of collaboration mode, the organic cooperation between humans and machines includes both simple collaboration where machine analysis comes first and human decision-making follows, and organic collaboration where algorithmic decision-making and human decision-making are deeply integrated. Third, from the perspective of capability evolution, the mutual achievement of human–machine collaboration includes both “intelligence augmentation” similar to the functioning of digital assistants, and human feedback reinforcement that helps digital employees “learn and grow.” In view of this, human–machine collaboration in such scenarios is no longer a simple tool-use process, but a partner interaction process in which, through human–machine knowledge sharing, deep coupling and mutual empowerment, both human autonomy and machine agency are emphasised—that is, “AI is an extension of human capability, and humans are amplifiers of AI capability”(12)—thereby evolving into a mature form of capitalist human–machine collaboration in the era of large-model production.

相比于前文所述的智能助手，数字员工在知识生产诸多环节的“劳动表现”已经与人类不相上下。正因如此，在数字资本推动的企业组织智能化变革中，智能机器开始以共同解决任务的“工作伙伴”的面貌出现，客观上对资本主义劳动过程产生了颠覆性影响。这种影响主要包括三个方面。一是从劳动地位来看，在人机团队协作中，数字员工虽然不能独立完成任务，但却承担着与人相当的工作量，自然不再是生产过程中的“配角”。二是从协作方式来看，人机之间的有机合作既有机器分析在先、人决策在后的简单协作，也有算法决策和人决策深度融合的有机协作。三是从能力演进来看，人机协作的相互成就既有类似数字助手作用发挥的“智能增强”,也有人反馈强化助力数字员工“学习成长”。鉴于此，此类场景中的人机协作不再是简单的工具使用过程，而是在人机知识共享、深度耦合和彼此赋能中既凸显人的自主性，又强调机器能动性的伙伴互动过程，即“AI是人的能力的延伸，人是AI能力的放大器”(12),进而演进成为大模型生产时代数字资本主义人机协作的成熟形态。

Third, general-purpose agents and human–machine collaboration. In the “automated machine system” that digital capital is striving to develop, a general-purpose agent is an intelligent machine capable of autonomously planning tasks, invoking tools and optimising pathways to complete specific tasks. At present, with the enhancement of large-model capabilities and the maturation of technologies such as retrieval-augmented generation (RAG), the ability of intelligent machine systems to perceive the environment, form memory, make decisions, take action and learn and optimise is also rapidly improving, iteratively developing into general-purpose agents and showing broad application prospects. From the perspective of the process of digital capitalist commodity production, on the one hand, in areas where intelligent machines excel, the emergence and self-iteration of general-purpose agents further promote the objectification of labour and the subjectification of machines in capitalist human–machine relations, which objectively, due to the accelerated “machine replacement of humans,” further impacts the production status of humans. On the other hand, in the “missing middle ground” where neither humans nor intelligent machines can complete work alone, general-purpose agents appearing as “tool users” will push digital capitalist human–machine collaboration into a new stage. A research report points out: “The emergence of AI agents means that the mode of collaboration between AI and humans will undergo a subversive change, shifting from the traditional ‘user-instruction-centred’ approach to a ‘user-intention-centred’ approach.”(13)

其三，通用智能体和人机协作。在数字资本倾力开发的“自动化机器体系”中，通用智能体是具备自主规划任务、调用工具、优化路径以完成特定任务的智能机器。当下，随着大模型能力的增强以及检索增强生成(RAG)等技术的成熟，智能机器系统感知环境、形成记忆、作出决策、采取行动和学习优化的能力也在快速提升，迭代发展成为通用智能体并显现出广阔的应用前景。从数字资本主义商品生产的过程来看，一方面，在智能机器擅长的领域，通用智能体的出场和自我迭代进一步推动了资本主义人机关系中的劳动客体化和机器主体化，这在客观上会因“机器换人”的提级加速而对人的生产地位产生进一步冲击。另一方面，在人和智能机器均无法独自完成工作的“缺失的中间地带”,以“工具使用者”的身份出现的通用智能体将推动数字资本主义人机协作迈向崭新的阶段。有研究报告指出：“AI智能体的出现意味着AI与人的协作方式将发生颠覆性变化，从传统的‘以用户指令为中心’,转变到‘以用户意图为中心’。”(13)

It can be foreseen that “the future work environment will be an ecosystem in which humans collaborate closely with highly specialised AI agents”.[14]14 Huang Zaisheng, “The Logic of Capital and the Future of Work in the Age of AI: A Marxist Critique of the ‘End of Work’ Thesis,” *Journal of Economic and Social Theory*, vol. 12, no. 3 (2024), p. 45. In human–machine collaboration, workers are only responsible for setting goals, providing resources and verifying results; the general-purpose agent directly faces the target task, breaks it down, selects tools, controls progress, and autonomously ends work once the goal is achieved. This means that, unlike digital employees who rely on human prompts to complete automation, the general-purpose agent achieves full automation simply through human goal-setting. Moreover, the general-purpose agent can achieve continuous evolution through interactive learning with humans, truly becoming the organisation’s highly intelligent digital avatar, thereby promoting intelligent collaboration in human–machine integration. In the “agent-as-a-service” launched by digital capital using artificial intelligence, the general-purpose agent, as an “auto-pilot”, perfectly acts as the human’s work proxy, with humans only responsible for setting target tasks and evaluating results. Of course, although workers gradually retreat to the background, they remain indispensable in the “reproduction of labour power” of the general-purpose agent through the “human-in-the-loop” mode.

可以预见，“未来的工作环境将是一个人与高度专业化的AI智能体紧密协作的生态”(14)。在人机协作中，劳动者只负责设定目标、提供资源和核验结果；通用智能体直接面对目标任务进行任务拆分、工具选择、进度控制，并在实现目标后自主结束工作。这意味着，不同于数字员工需借助人工提示词来完成自动化，通用智能体通过人设定目标就能全面实现自动化。此外，通用智能体能够通过与人进行互动学习来实现持续进化，真正成为组织的高度智能化数字分身，从而促进人机融合的智能协作。在数字资本推出的人工智能“代理即服务”(agent-as-a-service)中，通用智能体作为“主驾驶”(auto-pilot)完美地充当了人的工作代理，人只需负责设定目标任务和评估结果。当然，劳动者虽然逐渐隐身幕后，但依然以“人机回圈”的方式在通用智能体的“劳动力再生产”中不可或缺。

(III) The Dynamic Evolution of Human–Machine Collaboration under Large-Model Production

(三)大模型生产下人机协作的动态演化

Marx believed that “increasing the productivity of labour and negating necessary labour to the greatest possible extent … is the necessary tendency of capital.”[15]15 Karl Marx, *Grundrisse* (Penguin, 1973), p. 401. The current acceleration of the generalisation of intelligent machines’ capabilities makes a new round of “machines replacing humans” an irresistible trend, but for a considerable period, their level of autonomous intelligence will be far from sufficient to support the “full automation” of human knowledge production. Research shows that the functioning of intelligent machines’ “labour capacity” under large-model production exhibits a “jagged technological boundary”. Within the scope of AI’s capabilities, intelligent machines can efficiently and easily complete or even replace human work; but for tasks beyond their capability range, AI’s “performance” is mediocre, and deliberately using AI may actually lower human work levels.[16]16 See Ajay Agrawal, Joshua Gans, and Avi Goldfarb, *Prediction Machines: The Simple Economics of Artificial Intelligence* (Harvard Business Review Press, 2018), pp. 67–82. Precisely for this reason, as the intelligent production of digital capitalism advances, a “missing middle ground” will continuously emerge, requiring human–machine collaboration to fill it in a timely manner.

马克思认为：“提高劳动生产力和最大限度否定必要劳动……是资本的必然趋势。”(15)当下的智能机器通用性能泛化加速，新一轮“机器换人”的趋势势不可当，但在相当长一段时期内，其自主智能水平远不足以支撑人类知识生产“全自动化”。有研究表明，大模型生产下的智能机器“劳动能力”的作用发挥存在“锯齿状的技术边界”。在人工智能的能力范围之内，智能机器可以高效轻松地完成甚至取代人类工作；但面对其能力范围之外的任务，人工智能的“绩效表现”则一般，刻意使用人工智能反而会拉低人的工作水平。(16)正因如此，随着数字资本主义智能化生产的持续推进，“缺失的中间地带”会不断出现，这就需要人机协作加以适时填充。

Statically, the cognitive computing expertise and capability radius of intelligent machines directly determine the breadth and depth of human–machine collaboration. Dynamically, with the development of generative AI technology—high-quality production of AI training data, ample supply of advanced computing power, and continuous optimisation of algorithmic architectures—intelligent machines will continuously acquire human-like complex cognitive abilities, growing in more and more knowledge-work fields from “novices” into “digital experts” or even “human mentors”. The capability boundaries of intelligent machines constantly expand, thereby driving the accelerated development of human–machine collaboration. On the one hand, the “labour capacity” of intelligent machines significantly improves, so that more cognitive work tasks can be replaced by AI technology; tasks that previously required close cooperation between humans and machines can now be automatically executed by “digital employees”, and the established human–machine collaboration transforms into a new round of human–machine division of labour. On the other hand, in more complex cognitive decision-making fields, the “missing middle ground” that emerges will again stage the scene of “human–machine co-dancing”. From this perspective, the evolution of human–machine collaboration under large-model production is essentially a real reflection of the contradictory movement of the “automation–augmentation paradox”[17]17 The “automation–augmentation paradox” refers to the phenomenon whereby AI both replaces human labour (automation) and enhances human capabilities (augmentation), creating a contradictory dynamic in the labour process. in the application of intelligent machines under digital capitalism. It should be pointed out that, under the “Turing machine” paradigm of AI, human labour remains irreplaceable. With the development of large-model production, algorithmic decision-making and human decision-making in complex cognitive tasks become increasingly intertwined and inseparable, and human–machine symbiosis will become the most epochal production characteristic of the digital capitalist labour process.

静态地看，智能机器拥有的认知计算专长及其能力半径直接决定了人机协作的广度和深度。动态地看，随着生成式人工智能技术的发展，人工智能训练数据高质量生产、先进算力充沛供给、算法架构优化持续推进，智能机器将不断获取类人的复杂认知能力，在越来越多的知识工作领域从“初学者”成长为“数字专家”甚至“人类导师”。智能机器的能力边界不断拓展，进而推动了人机协作的加速发展。一方面，智能机器的“劳动能力”显著提升，有更多的认知型工作任务可以被人工智能技术所替代，原本需要人和机器通力合作完成的工作任务现在可以交由“数字员工”来自动执行，昔日成型的人机协作如今转变成为新一轮的人机分工。另一方面，在更加复杂的认知决策领域出现的“缺失的中间地带”又会重演“人机共舞”的画面。由此观之，大模型生产下的人机协作演化本质上是数字资本主义智能机器应用中“自动—增强悖论”(17)矛盾运动的现实反映。需要指出的是，在人工智能“图灵机”范式下，人的劳动始终不可替代。随着大模型生产的发展，复杂认知任务中算法决策与人决策日趋相互交织、密不可分，人机共生将成为数字资本主义劳动过程最具时代性的生产特征。

II. “Re-humanisation”: The New Discourse of Digital Capital Discipline

二、“再人性化”:数字资本规训的新话术

Entering the era of large-model production, digital capital embraces technological solutionism, fabricates a new AI fetish, and loses no time in loudly proclaiming that human–machine collaboration can “let people work like humans, not like machines”. In the view of AI enthusiasts, “AI can make human workers more human”.[18]18 This phrase is a common slogan in contemporary AI industry discourse, often attributed to tech executives promoting human-centred AI. Little do they know that Marx long ago said that within the capitalist system, “it is an extremely absurd bourgeois cliché to say that because the capitalist uses fixed capital (and fixed capital is itself the product of labour and merely the labour of others appropriated by capital) to lighten the worker’s labour (on the contrary, the capitalist uses machinery to strip labour of all independence and attractiveness) or to shorten the worker’s labour time, the worker therefore shares in the product of labour with the capitalist.”[19]19 Karl Marx, *Grundrisse* (Penguin, 1973), p. 701.

进入大模型生产时代，数字资本拥抱技术解决主义，炮制人工智能新物神，并不失时机地大肆宣扬人机协作可以“让人像人、而不是像机器一样地工作”。在人工智能的拥趸们看来，“人工智能可以让人类工作者变得更加人性化”(18)。殊不知，马克思早就说过，在资本主义制度内部，“说什么由于资本家利用固定资本(况且，固定资本本身就是劳动的产品，并且不过是被资本占有的他人劳动的产品)使工人的劳动减轻了(相反，资本家利用机器使工人的劳动失去了一切独立性和吸引力),或者使工人劳动的时间缩短了，所以工人就和资本家分享劳动产品了，这种说法是极其荒谬的资产阶级滥调”(19)。

(I) “Replacement of Repetitive Labour” and the “Re-humanisation” of Labour Content

(一)“重复性劳动替代”与劳动内容“再人性化”

As is well known, in capitalist mechanised mass production, workers, passively adapting to the rhythm of “machine time”, “do not freely develop their physical and mental energies but instead mortify their bodies and ruin their minds”.[20]20 Karl Marx, *Economic and Philosophic Manuscripts of 1844* (Progress Publishers, 1959), p. 30. With the development of large-model production, facing increasingly fierce digital labour resistance, digital capital, well-versed in survival tactics, strains to glorify AI, advocates human–machine reconciliation, and peddles the “new world of work” of human–machine collaboration everywhere. Therefore, the current market narrative of human–machine collaboration is actually a new trick by digital capital to re-enact the “manufacture of consent” by leveraging the AI revolution. In practice, digital capital deliberately creates a production illusion in which the “humanisation” of intelligent machines eliminates “dead labour dominating living labour”, while secretly accelerating the development of a new generation of AI technology, using the rapid iteration and application of large models to more frantically “devour living labour”. In view of this, the claim that human–machine collaboration promotes the “re-humanisation” of labour content is merely an ideological rhetoric by which digital capital deliberately beautifies and “naturalises” the capitalist application of intelligent machines.

众所周知，在资本主义机器化大生产中，工人在被动地适应“机器时间”生产的节奏中“不是自由地发挥自己的体力和智力，而是使自己的肉体受折磨、精神遭摧残”(20)。随着大模型生产的发展，面对日趋激烈的数字劳动反抗，深谙生存之道的数字资本竭力溢美人工智能，宣扬人机和解，并四处兜售人机协作的“工作新世界”。因此，当下人机协作的市场叙事其实是数字资本借力人工智能革命来重施“同意的生产”的一种新伎俩。在实践中，数字资本蓄意制造出一种以智能机器的“人化”来消弭“死劳动支配活劳动”的生产幻象，暗地里却加紧开发新一代人工智能技术，以大模型的加速迭代和应用来更加疯狂地“吞噬活劳动”。鉴于此，所谓人机协作推动劳动内容“再人性化”的说法只不过是数字资本刻意美化“自然化”智能机器资本主义应用的一种意识形态说辞而已。

First, under the capitalist mode of production, “labour is external to the worker”.[21]21 Karl Marx, *Economic and Philosophic Manuscripts of 1844* (Progress Publishers, 1959), p. 31. For the proletarian trapped in labour alienation, no matter how light the work tasks in the production process of capital, “labour is always an odious thing, always appears as external forced labour”.[22]22 Karl Marx, *Grundrisse* (Penguin, 1973), p. 611. Moreover, in the human–machine opposition where the machine system becomes the subject and the worker becomes the tool, the capitalist labour process “suppresses the worker’s manifold productive inclinations and talents”.[23]23 Karl Marx, *Capital*, vol. 1 (Penguin, 1976), p. 548. Precisely for this reason, in the standardised assembly workshop or the informatised cubicle space, “the worker feels at home only outside labour, and in labour he feels homeless; he feels at ease when he is not working, and when he is working he feels ill at ease.”[24]24 Karl Marx, *Economic and Philosophic Manuscripts of 1844* (Progress Publishers, 1959), p. 31. There is no doubt that, in social life, as long as the rule of capital exists for a single day, labour will never be a “free expression of life” that brings people a humanly fulfilling joy of life. Entering the era of large-model production, one basic production fact that has not changed is that the “quality” of the digital capitalist labour process being subordinate to capital has not undergone any change. In the human–machine collaboration loudly promoted by digital capital, even if workers can engage in more challenging work tasks, it remains “a means of satisfying needs other than labour”, and thus inevitably remains a “self-sacrifice, self-torture of labour” subject to the will of capital.[25]25 Karl Marx, *Economic and Philosophic Manuscripts of 1844* (Progress Publishers, 1959), p. 32.

首先，在资本主义生产方式下，“劳动对工人来说是外在的东西”(21)。对于身陷劳动异化的无产者而言，在资本的生产过程中无论承担多么轻松的工作任务，“劳动始终是令人厌恶的事情，始终表现为外在的强制劳动”(22)。并且，在机器体系主体化与劳动者工具化的人机对立中，资本主义劳动过程“压抑工人的多种多样的生产志趣和生产才能”(23)。正因如此，在标准化的装配车间抑或信息化的格子空间里，“工人只有在劳动之外才感到自在，而在劳动中则感到不自在，他在不劳动时觉得舒畅，而在劳动时就觉得不舒畅”(24)。毋庸置疑，在社会生活中，只要资本统治还存在一天，劳动就不会是能给人带来合乎人性的生活乐趣的“自由的生命表现”。进入大模型生产时代，没有改变的一个基本生产事实是，数字资本主义劳动过程从属于资本的“质”并未发生任何改变。在数字资本高调推行的人机协作中，即使劳动者能够从事更具挑战性的工作任务，也依然是“满足劳动以外的那些需要的一种手段”,也就不可避免地依旧是服从资本意志的“一种自我牺牲、自我折磨的劳动”。(25)

Second, as Marx pointed out, every machine is “an organ of the human brain created by the human hand; the power of knowledge, objectified” (26). In the era of large-model production, the development and iteration of intelligent machines are no exception, and are even inseparable from human labor input at every moment. Thus, a highly ironic paradox emerges: digital capital spares no effort to develop intelligent machines that accelerate the replacement of repetitive labor, yet simultaneously it must scour the globe for cheap labor to complete digital tasks that, in the development, inference, deployment, and iteration of large models, cannot yet be encoded. This means that digital capital, on the one hand, loudly proclaims that artificial intelligence will liberate people from repetitive labor, while on the other hand, it constantly creates digital labor that is even more fragmented and devoid of labor content. In reality, “behind search engines, applications, and smart devices are vast numbers of workers, banished to the margins of the globalized system, lacking other options, forced to clean data and monitor algorithms, earning only a few cents in return” (27). What is more, in order to sustain the market narrative of AI as a “savior,” digital capital devises various strategies of “labor concealment,” turning countless anonymous online digital workers into lines of code within intelligent machine systems, thereby de-laboring their relations and forcing them into precarious, exhausting “ghost work.” On this point, one scholar pointedly remarks that the currently flourishing human-machine collaboration is in fact nothing but alienated and disguised interpersonal cooperation manipulated by digital capital (28).

其次，正如马克思所指出的，任何机器都是“人的手创造出来的人脑的器官；是对象化的知识力量”(26)。进入大模型生产时代，智能机器的开发与迭代也概莫能外，甚至是须臾离不开人的劳动投入。于是，一个极具讽刺意味的悖谬是，数字资本倾力开发加速替代重复性劳动的智能机器，但同时却又需要在全球范围网罗廉价劳动力来完成在大模型开发、推理、部署和迭代中一时还难以编码化的数字化任务。这意味着，数字资本一边高调宣扬人工智能将人从重复性劳动中解放出来，一边却又时刻在制造更加碎片化、更无劳动内容的数字劳动。现实中，“在搜索引擎、应用程序和智能设备的背后，是大量的工人，他们被放逐在全球化体系的边缘，缺乏其他选择，被迫清理数据和监督算法，仅能换得几美分的报酬”(27)。更有甚者，数字资本为维系人工智能“救世”的市场叙事，便想方设法地推行各种“劳动隐化”策略，将众多匿名在线的数字劳动者设置成为智能机器系统中的一行行代码，进而以去劳动关系化迫使其从事朝不保夕、身心俱疲的“幽灵工作”。对此，有学者一针见血地指出，时下勃兴的人机协作实际上是数字资本操纵下被异化和伪装的人际合作罢了。(28)

Furthermore, from the perspective of the species’ generic characteristics, humans excel at critical thinking and possess intuition, judgment, and empathy, enabling them to act opportunely in unpredictable scenarios. But at the individual level, people’s labor capacities vary. In reality, only a minority can fully realize their creative potential. For this reason, for ordinary workers, the image of intelligent machines performing “drudgery” while humans specialize in innovation remains merely a rosy work fantasy. The real situation is that large-model production is “liberating” humans from monotonous, repetitive work at the cost of massive job losses among ordinary white-collar workers. More notably, relevant empirical research shows that since the release of large models like ChatGPT, the cloud-based collaborative deployment of large models has indeed provided people with dazzling “intelligent tools,” but in terms of the actual effect of “intelligence augmentation,” human-machine collaboration does not necessarily bring digital workers the imagined ease and relaxation (29).

再者，从物种的类特质来看，人擅于批判性思考，拥有直觉力、判断力和同理心，因而在不可预测的场景中可以见机行事。但从个体来看，人的劳动能力有高有低。现实中，能够将创造潜力充分发挥出来的人毕竟只是少数。正因如此，对于普通劳动者而言，智能机器承担“苦役”、人专事于创新的画面只能算是一种美好的工作憧憬而已。真实的情形是，大模型生产正在以大量普通白领失去工作为代价将人类从单调重复的工作中“解放”出来。更值得一提的是，相关实证研究表明，自ChatGPT等大模型发布以来，大模型云端协同部署的确给人们带来了令人眼花缭乱的“智能工具”,但从其“智能增强”的实际功效来看，人机协作并不必然给数字劳动者带来想象中的工作自如和轻松。(29)

(II) “Increase in Free Time” and the “Re-Humanization” of Labor Methods

(二)“自由时间增加”与劳动方式“再人性化”

In the context of political economy, free time refers to leisure time, or time for engaging in higher-order activities. In the era of large-model production, digital capital vigorously promotes the superhuman potential of AI, tirelessly depicting an alluring vision of an intelligent society. In this society, intelligent machines tirelessly and efficiently produce, social wealth is vastly abundant, and humans have ample disposable free time—the life Marx described, where one “hunts in the morning, fishes in the afternoon, rears cattle in the evening, criticizes after dinner” (30), seems just around the corner. If so, why wouldn’t people actively embrace artificial intelligence? Yet this is merely a wishful practical fantasy of technological solutionism, untenable in theoretical terms.

在政治经济学语境中，自由时间是指闲暇时间，抑或从事高级活动的时间。进入大模型生产时代，数字资本极力宣扬人工智能的超人潜能，不厌其烦地描绘令人向往的智能社会图景。在这种社会，智能机器不知疲倦地高效生产，社会物产极大丰裕，人类拥有充足的可自由支配的时间，马克思笔下的“随自己的兴趣今天干这事，明天干那事，上午打猎，下午捕鱼，傍晚从事畜牧，晚饭后从事批判”(30)的生活似乎指日可待。若是如此，人们还有什么理由不去主动拥抱人工智能呢?然而，这其实是技术解决主义的一厢情愿式的实践臆想，在学理上实在不堪一驳。

First, from the production logic of capital, capital spares no effort to promote technological progress, using the machine system to extract relative surplus value to the greatest extent, objectively reducing the necessary direct labor input in the production process to a minimum. Thus, unintentionally, “capital thus works against its own will, creating the conditions for socially disposable free time” (31). However, value-based social production under the rule of capital will never allow workers to achieve their own development by acquiring free time. On the contrary, the capitalist application of the machine system always tends “on the one hand to create disposable free time, and on the other to convert that disposable free time into surplus labor” (32). In the era of large-model production, as digital capitalism’s generative AI technology accelerates, the “higher-order labor capacity” acquired by intelligent machines continuously improves, and the “thorough replacement of living labor by dead labor” is vigorously advanced. Faced with this crisis of value movement, in order to sustain value-based production under large-model production conditions, digital capital’s greed for surplus labor will become more urgent than at any time since the First Industrial Revolution.

首先，从资本的生产逻辑来看，资本不遗余力地推动技术进步，利用机器体系来最大程度地榨取相对剩余价值，客观上使生产过程中直接劳动的必要投入缩减到最低程度。进而，无意间，“资本就违背自己的意志，成了为社会可以自由支配的时间创造条件的工具”(31)。但是，资本统治下的以价值为基础的社会生产绝不会任由劳动者通过获取自由时间来实现自身的发展。相反，机器体系资本主义应用的趋势始终是“一方面创造可以自由支配的时间，另一方面把这些可以自由支配的时间变为剩余劳动”(32)。进入大模型生产时代，随着数字资本主义生成式人工智能技术的加速发展，智能机器获得的“高阶劳动能力”持续提升，“死劳动对活劳动的彻底替代”得到大力推进。面对这一价值运动危机，为了维系大模型生产条件下以价值为基础的生产，数字资本对剩余劳动的贪求将会比第一次工业革命以来的任何时候都急迫。

Second, looking back at the history of capitalist development, no matter how far social productive forces have developed or to what extent they have created “conditions for the liberation of labor,” under the rule of capital, free time has always been the non-labor time of a minority. The capitalist who steals the worker’s surplus labor time is the true master of time; all his time is free, or even his necessary labor time is his free time. In contrast, “the proletarian is merely a machine for the production of surplus value” (33). The triumphant advance of intelligent machine application in digital capitalist production reduces the total labor time of society at an ever more astonishing acceleration, thereby providing unprecedented material and technical preparation for the liberation of labor. But in reality, digital capital’s consolidation of platform monopolies, imposition of algorithmic tyranny, and pursuit of AI monetization—all these profit-driven operations—unfailingly demonstrate that digital capitalism’s social production, which “takes labor time as the measure of wealth,” will not only not recede but, in the process of promoting “the digitization of the real world and the materialization of the digital world,” will more wantonly stage its old trick of “dead labor dominating living labor.” From this it can be inferred that the human-machine collaboration under digital capitalist large-model production will likely bring the increasingly platform-dependent masses even more “leisureless” digital busyness and even deeper “time poverty.” Recent foreign research shows that workers with higher occupational exposure to AI generally face longer working hours and reduced leisure time (34).

其次，纵观资本主义发展历程，不管社会生产力发展到何种程度、在多大程度上创造了“劳动获得解放的条件”,在资本统治下的社会生活中，自由时间都只是少数人的非劳动时间。偷窃工人剩余劳动时间的资本家是时间的真正主人，他的所有时间都是自由的，或者说他的必要劳动时间也是其自由时间。相比之下，“无产者不过是生产剩余价值的机器”(33)。数字资本主义生产中智能机器应用的高歌猛进以更加令人震撼的加速度缩减了整个社会的劳动时间，从而能够史无前例地为劳动解放提供物质技术准备。但在现实中，数字资本巩固平台垄断，推行算法专制，谋求人工智能变现，凡此种种的逐利操作无一不在昭示：数字资本主义“以劳动时间作为财富的尺度”的社会生产不仅不会消退，反而在推进“现实世界数字化、数字世界实体化”的过程中更加恣意地上演着“死劳动支配活劳动”的拿手好戏。由此推之，数字资本主义大模型生产下的人机协作给日趋依附平台的社会大众带来的恐怕将是更加“无问闲暇”的数字忙碌和更加深重的“时间贫困”。国外近期的相关研究显示，职业人工智能暴露程度较高的劳动者普遍面临工作时间延长和休闲时间减少的困境。(34)

(III) Increase in Labor Productivity and the “Re-Humanization” of Labor Outcomes

(三)劳动生产率提升与劳动结果“再人性化”

Entering the era of large-model production, the intelligent machines that capital is competing to develop have promoted the industrialization of knowledge commodity production and the “intelligent enhancement” of laborers, and their potential to promote the development of productive forces should not be underestimated. Relevant empirical research results show that in fields of knowledge commodity production such as copywriting, customer service, creative production, and code development, human-machine collaboration under large-model production has driven a significant increase in labor productivity.[35]35 See, for example, Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, “Generative AI at Work,” NBER Working Paper, No. 31161, 2023; Shakked Noy and Whitney Zhang, “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence,” Science, Vol. 381, No. 6654, 2023, pp. 187–192. In some people’s eyes, large-model production possesses a natural technological inclusiveness, and the “digital dividend” brought by human-machine collaboration will benefit every new type of laborer who embraces artificial intelligence. Following this logic, the problem of human alienation from one’s own labor product, which has long been caused by the separation of laborers from the means of production, will be resolved in the human-machine co-creation of the large-model production era. In fact, Marx profoundly pointed out that within the capitalist system, “all means for developing the productivity of labor are brought about at the cost of the individual worker.”[36]36 Karl Marx, Capital, Vol. 1, Beijing: People’s Publishing House, 2004, p. 743. Besides “degrading the worker to the level of an appendage of a machine” and “transforming the worker’s lifetime into working-time,”[37]37 Karl Marx, Capital, Vol. 1, p. 743. the capitalist application of machinery “turns the producer into a pauper.”[38]38 Karl Marx, Capital, Vol. 1, p. 508. Looking back at the history of capitalist labor, from the industrial workers of the machine-based large-industry era to the knowledge workers of the information age, even though “labor power may, through class struggle, raise wages and improve living conditions,” a basic fact of capitalist national income distribution is that “the benefits that machines could bring us in the past are actually the benefits that machines could bring to capitalist enterprises.”[39]39 Yao Shunqi, “The Political Economy of Artificial Intelligence,” Social Sciences in China, No. 9, 2021, p. 117. Entering the digital capitalist stage, the digital piecework wage system implemented by capital has trapped numerous “clickworkers” and informal laborers in the exploitation of digital sweatshops, increasingly falling into the dual survival predicament of “employment instability” and “data poverty.”

进入大模型生产时代，资本竞相开发的智能机器推动了知识商品生产的工业化和劳动者的“智能增强”,其在促进生产力发展方面的潜能不可小觑。相关实证研究结果显示，在文案撰写、客户服务、创意生产、代码开发等知识商品生产领域，大模型生产下的人机协作推动了劳动生产率的显著提升。(35)在一些人眼中，大模型生产具有天然的技术普惠性，人机协作带来的“数字红利”将会惠泽、拥抱人工智能的每一名新型劳动者。照此逻辑推之，一直以来因劳动者与生产资料相分离而导致的人与自己的劳动产品相异化的问题，在大模型生产时代的人机共创中将得以消解。 而事实上，马克思曾深刻地指出，在资本主义制度内部，“一切提高社会劳动生产力的方法都是靠牺牲工人个人来实现的”(36)。除了“把工人贬低为机器的附属品”“把工人的生活时间转化为劳动时间”(37),机器的资本主义应用“使生产者变成需要救济的贫民”(38)。回顾资本主义劳工史，从机器大工业时代的产业工人到信息时代的知识工人，纵然“劳动力可能通过阶段斗争来提高工资和改善生活条件”,但资本主义国民收入分配的一个基本事实是，“过去机器能够为我们带来的好处其实是机器能够为资本主义企业带来的好处”。(39)进入数字资本主义阶段，资本推行的数字计件工资制使众多“键盘点击工”(clickworker)和非正规劳工身陷数字血汗工厂的盘剥，日益陷入“就业不稳定”和“数据贫困”的双重生存窘境。

At present, capitalist platforms are becoming increasingly entrenched, digital capital monopolizes key digital means of production, and laborers’ substantive dependence on platforms is deepening day by day. Faced with the comprehensive domination of digital capital, people cannot help but ask themselves: How can merely skillfully operating various intelligent tools fulfill the desire to “possess one’s own surplus labor”? Analyzing the current distribution situation in contemporary capitalist society, an increasingly obvious fact is: “The capitalist application of artificial intelligence, on the one hand, creates极致生产力 [extreme productive forces], and on the other hand, these productive forces are not commonly possessed by the whole society, but are possessed by the capitalist class.”[40]40 Yao Shunqi, “The Political Economy of Artificial Intelligence,” p. 117. Consequently, the market dominance and the huge profits seized by a handful of digital technology giants are rare in the world, enough to make the plundering acts of the capitalist “robber barons” of the late 19th century pale in comparison.

当下，资本主义平台愈发牢固，数字资本垄断关键数字生产资料，劳动者对平台的实质性依附日趋加深。面对数字资本的全面宰制，人们不禁自问：仅仅熟练操作各种智能工具何以如愿地实现“占有自己的剩余劳动”?剖析当代资本主义社会的分配现状，一个愈发明显的事实是：“人工智能的资本主义应用一方面创造了极致生产力，另一方面这种生产力不是由全社会共同占有，而是由资本家阶级占有。”(40)于是，屈指可数的数字科技巨头占据的市场支配地位及其攫取的高额利润举世罕见，足以使19世纪末资本主义“强盗大亨”们的攫取行径相形见绌。

III. Alternative “Deskilling”: New Manifestations of the Degradation of Human Labor

三、另类“去技能化”:人类劳动退化的新表征

Fundamentally, the core of the artificial intelligence revolution is “replacing human mental labor with artificial intelligence systems.”[41]41 He Zhaotian, “The Logic of Capital and the Logic of Labor in the Era of Artificial Intelligence,” Journal of Renmin University of China, No. 6, 2021, p. 62. Entering the era of large-model production, digital capital, on the one hand, is busy painting human-machine collaboration with the “rosy hue” of “re-humanization,” while on the other hand, it secretly intensifies the use of “machines replacing humans” to increase profits. The result is that in the deepening human-machine division of labor, the large-scale application of intelligent machines under digital capitalism accelerates the “deskilling” of knowledge commodity production. At the same time, in the increasingly expanding human-machine collaboration, “intelligent enhancement” and excessive reliance on artificial intelligence have triggered a “deskilling” of the capitalist labor process different from before.

从根本上讲，人工智能革命的“核心是以人工智能系统替代人类的脑力劳动”(41)。进入大模型生产时代，数字资本一边忙于给人机协作抹上“再人性化”的“玫瑰色”,一边暗地里加紧通过“机器换人”来提高收益。其结果是，在不断深化的人机分工中，数字资本主义智能机器的规模化应用加速推动了知识商品生产的“去技能化”。与此同时，在日趋扩大的人机协作中，“智能增强”和人工智能过度依赖也引发了不同于以往的资本主义劳动过程“去技能化”。

(1) Intelligent Substitution and the Accelerated Devaluation of Intellectual Labor Skills

(一)智能化替代与智力劳动技能的加速贬值

Labor process theory holds that a production tenet capital never forgets is that “the labor process should not rely on craftsmanship, tradition, and the worker’s knowledge.”[42]42 Harry Braverman, Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century, trans. Fang Sheng et al., Beijing: The Commercial Press, 1978, p. 111. Since the First Industrial Revolution, capital has continuously promoted the “deskilling” of workers’ labor methods through division of labor and expanding the scale of division of labor. Entering the digital capitalist stage, the emergence of intelligent machines has driven capital to decompose, reorganize, and replicate human interpretive capabilities throughout the entire production process.[43]43 See Nick Dyer-Witheford, Atle Mikkola Kjøsen, and James Steinhoff, Inhuman Power: Artificial Intelligence and the Future of Capitalism, London: Pluto Press, 2019, p. 43. With the support of generative artificial intelligence technology, digital capital steadily promotes the machine substitution of human cognitive tasks. At the same time, platform enterprises build digital factories, launch a “human-as-a-service” model, and create a large amount of “abundant, calculable, and easily replaceable” digital labor.

劳动过程理论认为，资本念念不忘的一个生产信条就是“劳动过程不应该依靠手艺、传统和工人的知识”(42)。自第一次工业革命以来，资本通过分工和扩大分工规模来不断推进工人劳动方式的“去技能化”。进入数字资本主义阶段，智能机器的出场推动了资本对整个生产过程中人的解释能力进行分解、重组和复制。(43)在生成式人工智能技术的加持下，数字资本稳步推动对人类认知任务的机器替代。与此同时，平台企业搭建数字工厂，推出“人工即服务”(human-as-a-service)模式，制造出大量“供给充足、可计算和易于替代”的数字劳动。

Entering the era of large-model production, the speed of technological iteration of generative artificial intelligence is astonishing, and intelligent machines are increasingly acquiring human-like characteristics of “strong intelligence, strong generality, and strong interactivity.” In practice, digital capital vigorously develops the decomposition, replication, and imitation of the labor process of knowledge commodity production, and intensifies a new round of “machines replacing humans,” that is, accelerating the transfer of organizational knowledge and expert capabilities into intelligent machine systems. Under the influence of the application of artificial intelligence technology, on the one hand, human intellectual labor capabilities are continuously “devalued” or even “have no place to use”; on the other hand, the content of digital labor serving large-model production is simplified, and the mechanization of labor methods is intensified. It can be said that digital capital is pushing the “deskilling” of the capitalist labor process to a new height.

进入大模型生产时代，生成式人工智能的技术迭代速度令人瞠目，智能机器日益获得了“强智能性、强通用性、强交互性”的类人特征。在实践中，数字资本大力发展知识商品生产劳动过程的拆解、复制和模仿，并加紧推进新一轮“机器换人”,也即将组织知识和专家能力加速转移到智能化机器体系之中。在人工智能技术应用的影响下，一方面，人类的智力劳动能力不断“贬值”甚至“无用武之地”;另一方面，服务于大模型生产的数字劳动内容简单化、劳动方式机械化程度加剧。可以说，数字资本正在将资本主义劳动过程的“去技能化”推向新高度。

(2) Intelligent Enhancement and the Dissolution of the “Skill Premium” for Skilled Labor

(二)智能增强与熟练劳动“技能溢价”的消解

During the period of industrial capitalism, machine-based large-scale production made workers’ labor increasingly simplified, one-sided, and mechanized. From the perspective of the evolution of labor forms, “in the automatic factory, what replaces the hierarchy of specialized workers characteristic of manufacture is the tendency towards the equalization or homogenization of the various tasks performed by the assistants of the machine.”[44]44 Karl Marx, Capital, Vol. 1, p. 484. Different from the homogenization of labor content, from the perspective of the application effect of intelligent assistants, human-machine collaboration under large-model production continuously releases the productive potential of “intelligent enhancement,” greatly eliminating differences in labor skills, so that the labor performance of different employees is no longer far apart, but presents a “performance leveling” effect. For low-skilled employees with shallow learning, the emergence of intelligent tools is like a timely help, enabling them to quickly load expert knowledge and obtain specific guidance, thereby significantly improving labor productivity. In contrast, for high-performing employees with rich work experience, the assistance of intelligent tools is only the icing on the cake, with little benefit to their performance improvement. A survey of 5,179 customer agents at a call center in the Philippines showed that after using an intelligent assistant software, the work efficiency (measured by the number of problems solved per hour) of these employees engaged in customer support work increased by an average of 14%. However, this study also found that the work efficiency of new employees and low-skilled employees improved particularly significantly. In contrast, the empowering effect of generative artificial intelligence on experienced employees was not obvious.[45]45 See Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, “Generative AI at Work,” NBER Working Paper, No. 31161, 2023.

在工业资本主义时期，机器化大生产使工人劳动愈加简单化、片面化和机械化。从劳动形态演进来看，“在自动工厂里，代替工场手工业所特有的专业化工人的等级制度的，是机器的助手所要完成的各种劳动的平等化或均等化的趋势”(44)。不同于劳动内容的趋同化，从智能助手的应用功效来看，大模型生产条件下的人机协作不断释放“智能增强”的生产力潜能，极大地消弭了劳动技能差异，使不同员工的劳动表现不再相差很远，而是呈现出“业绩拉平”的效果。对于那些学艺尚浅的低技能员工，智能工具的出现如同雪中送炭，使其能够迅速加载专家知识、获取具体指导，从而显著地提升劳动生产率。相比之下，对那些拥有丰富工作经验的高绩效员工来说，智能工具的助力只能算是锦上添花，对其绩效的提高助益不大。对菲律宾一家呼叫中心的5179名客户代理的调查表明，在使用一款智能助手软件之后，这些从事客户支持工作的员工的工作效率(以每小时解决问题的数量来衡量)平均提高了14%。但是，这一研究也同时发现，新员工和低技能员工的工作效率的提高尤其显著。相比之下，生成式人工智能对经验丰富的员工的赋能作用并不明显。(45)

In short, the intelligent enhancement and “productivity equalizer effect” brought by large-model production have, to a certain extent, reversed the “skill-biased technological change,” promoting the convergence of labor skills between skilled and unskilled workers within the organization. This means that the application of intelligent tools can effectively lower the technical threshold for knowledge commodity production, weaken the competitive advantage brought by workplace work skills, and narrow the performance gap between new and old employees. As a result, the “skill premium” previously enjoyed by skilled workers is greatly dissolved,[46]46 “Skill premium” refers to the wage premium that skilled workers receive over unskilled workers due to their higher productivity. triggering a new round of “deskilling” of knowledge workers in the sense of labor compensation. Objectively speaking, the “deskilling” of skilled workers in this situation helps to narrow the pay gap within the organization. However, from the perspective of the development trend of the working class composition, the “intelligent enhancement” achievable by everyone brought by artificial intelligence will make former technical backbones and senior white-collar workers find it difficult to realize their “middle-class dream” due to the devaluation of “human capital,” thus giving rise to “a huge new type of proletariat.”[47]47 Yao Shunqi, “The Political Economy of Artificial Intelligence,” p. 118.

简而言之，大模型生产带来的智能增强和“生产力均衡器效应”在一定程度上逆转了“技能偏向型技术变迁”,推动了组织内熟练工人与非熟练工人劳动技能的趋同化。这意味着，智能工具的应用可以有效地降低知识商品生产的技术门槛，削弱职场工作技能带来的竞争优势，拉近新老员工的业绩表现。如此一来，原先熟练工人享有的“技能溢价”被极大地消解了(46),这在劳动补偿的意义上引发了新一轮知识工人的“去技能化”。客观上说，这种情境下熟练工人的“去技能化”有助于缩小组织内的薪酬差距。然而，从工人阶级构成的发展趋势来看，人工智能带来的人人旨可实现的“智能增强”将会使曾经的技术骨干和高级白领因“人力资本”的贬值而难圆“中产梦”,从而出现“一支庞大的新型无产阶级”(47)。

(3) The Generative Paradox of Human-Machine Cooperation and Complex Labor Capabilities

(三)人机合作与复杂劳动能力的生成悖论

From the perspective of the application scenarios of digital employees, in the process of human–machine collaboration, intelligent machines exert their powerful perceptual and computational capabilities to complete “cognitive decision-making” in routine scenarios; humans, by contrast, use their own imagination, intuition, and judgment to undertake random decision-making tasks under uncertain circumstances. In the short term, human–machine collaboration will bring digital capital considerable surplus value and reduce the workload of knowledge workers. In the long run, however, this human–machine collaboration inadvertently severs the generative chain of human complex cognitive abilities, creating a “knowledge acquisition paradox” [48]48 The “knowledge acquisition paradox” refers to the situation where reliance on AI for routine cognitive tasks prevents workers from developing the complex cognitive skills needed for non-routine decisions., thereby inducing a de-skilling in the genetic sense. Moreover, “as computers begin to occupy more and more knowledge-work tasks, the speed of skill degradation will accelerate” [49]49 This is a citation from a source not specified in the text.. This is because the generation and enhancement of human cognitive abilities do not occur in a vacuum but arise from concrete, rich and varied social practice. For ordinary workers, only by undergoing job training at different levels and “learning by doing”, continuously acquiring and consolidating the “basic skills” of intellectual labour, can they “polish a rough jade into a finished vessel” and eventually become experts with specialist competence. If workers are intoxicated with having digital employees “take over” routine decision-making and procedural work, so that organisational employees are deprived of necessary job training and cannot even form the most basic business understanding or work intuition, then an unavoidable practical problem arises: where do the complex cognitive abilities that workers must possess in human–machine collaboration—such as cognitive judgment and on-the-spot handling—come from? The latest empirical study shows that “cognitive offloading” caused by frequent use of artificial intelligence tools is detrimental to the generation of human critical thinking ability [50]50 This is a citation from a source not specified in the text.. As intelligent machines accelerate their evolution toward general intelligent agents, in digital capitalist large-model production, on the one hand, human intellectual labour input is reduced to a minimum; on the other hand, higher-order cognitive abilities such as human strategic conception and situational judgment become increasingly indispensable. If this trend continues, the widespread adoption of human–machine collaboration in digital capitalist large-model production will make the paradox of generating human complex labour capacity ever more intractable and acute.

从数字员工的应用场景来看，在人机协作的过程中，智能机器发挥自身强大的感知、运算能力，完成常规场景的“认知决策”;人则利用自身的想象力、直觉力和判断力，承担不确定情形下的随机决策任务。短期内，人机协作将会给数字资本带来可观的超额剩余价值，并减轻知识劳动者的工作负荷。但从长远来看，这种人机协作无意间也阻断了人类复杂认知能力的生成链路，制造了“知识获取悖论”(48),从而引致发生学意义上的“去技能化”。此外，“随着计算机开始占据越来越多的知识工作任务，技能退化的速度将会加快”(49)。这是因为，人类认知能力的生成与提高不是在真空中发生的，而是产生于具体的、丰富多彩的社会实践。对于普通劳动者而言，只有经历不同层级的岗位磨砺和“干中学”,不断获取和夯实智力劳动“基本功”,才有可能“琢玉成器”,最终成为具备专家能力的行家里手。如果一味地陶醉于由数字员工“代劳”常规决策和程序性工作，使组织员工得不到必要的岗位历炼，甚至连最基本的业务理解和工作直觉都无法形成，那么一个不容回避的实践难题是，人机协作中劳动者所必须具备的认知决断、临场处置等人的复杂认知能力从何而来?最新的一项实证研究表明，经常使用人工智能工具而导致的“认知卸载”(cognitive offloading)不利于人类批判性思维能力的生成。(50)随着智能机器朝向通用智能体加速演化，在数字资本主义大模型生产中，一方面，人类的智力劳动投入缩减到最低限度；另一方面，人类的战略构想、情境决断等高阶认知能力将变得愈发不可或缺。照此情形发展下去，数字资本主义大模型生产中的人机协作大行其道，人类复杂劳动能力的生成悖论将变得愈加棘手和尖锐。

(4) Excessive dependence on artificial intelligence and the blunting of human creative ability

(四)对人工智能的过度依赖与人类创造能力的钝化

For a long time, humans have harboured an inherent fascination with “the automatic action of machines” [51]51 This is a citation from a source not specified in the text.. Entering the digital capitalist stage, the capitalisation of technology and the technologisation of capital have accelerated, and intelligent machines represented by search engines and recommendation algorithms have had a significant impact on human production and life. Research shows that, faced with increasingly sophisticated intelligent machine systems, people easily fall into the psychological quagmire of over-trusting machines, producing “automation bias” [52]52 “Automation bias” is a psychological tendency to trust automated systems over human judgment.. Compared with human decision-making, people believe machine decision-making is more efficient, accurate, and objective. With the accelerated deployment of large-model cloud-based collaboration, intelligent machines learn common knowledge from massive data, achieving “the ability to think, create, make decisions, and reason” through algorithms with strong generalisation, deep structure, and self-adaptation, increasingly demonstrating powerful AI content-generation capabilities. From the perspective of the development of human–machine relations, people’s infatuation with and excessive dependence on digital capitalist intelligent machines is deepening.

一直以来，人类对“机器的自动作用”具有内在迷恋。(51)进入数字资本主义阶段，资本技术化和技术资本化得以加速推进，以搜索引擎、推荐算法等为代表的智能机器对人类生产生活产生了重要影响。有研究表明，面对日趋精进的智能化机器体系，人们容易陷入过度信任机器的心理泥沼，产生“自动化偏见”(automation bias)。(52)相较于人工决策，人们认为机器决策更加高效、准确和客观。随着大模型云端协同部署的加快推进，智能机器从海量数据中学习共性知识，通过强泛化、深结构和自适应的算法处理达到“能思考、做创作、善决策、会推理”的目的，日趋展现出强大的人工智能内容生成能力。从人机关系发展来看，人们对数字资本主义智能机器的痴迷和过度依赖有加深之势。

At present, generative AI technology is advancing rapidly, inaugurating an industrial revolution in the field of capitalist knowledge commodity production. In the process of chasing surplus value, the development and production of creative commodities are increasingly relied upon by digital capital, and the scarcity of human imagination and creativity as factors is thus becoming more prominent. At the same time, however, because large models increasingly exhibit unexpected “creativity”, especially shining in drug research and development and scientific research, people’s dependence on AI in innovative work is also beginning to emerge. On the surface, in the production of creative commodities, humans “refining” on the basis of the “first draft” completed by intelligent machines appears to yield twice the result with half the effort, greatly improving innovation efficiency. In reality, however, only employees with high work skills can benefit from collaboration with AI and enhance their creativity [53]53 This is a citation from a source not specified in the text.; moreover, in the long run, when we regard AI as an “innovator”, human–machine collaboration may ultimately bring about the annihilation of human creativity and originality [54]54 This is a citation from a source not specified in the text.. An experiment on creative writing shows that although human–machine collaboration can enhance individual creativity to a certain extent, the novelty of collective output is impaired [55]55 This is a citation from a source not specified in the text..

目前，生成式人工智能技术狂飙突进，正在开启资本主义知识商品生产领域的工业革命。在追逐超额剩余价值的过程中，创意类商品的开发与生产愈发被数字资本所倚重，人的想象力和创造力的要素稀缺性因此也日趋凸显。但与此同时，由于大模型日趋展现出出乎人意料的“创造性”,尤其是在药物研发和科学研究中大放异彩，人们在创新性工作领域的人工智能依赖也初露端倪。表面上看，在创意类商品生产中，人类在智能机器完成的“初稿”基础上进行“精加工”会显得事半功倍，极大地提高创新效率。但实际上，不仅实践中只有具备高工作技能的员工才能在与人工智能的协作中受益，提高创造力；(53)而且从长远计，当我们将人工智能视同“创新者”时，人机协作最终带来的可能是人类创造力和原创性的湮灭。(54)一项针对创意写作的实验表明，尽管人机协作能够在一定程度上提升个体的创造力，但集体输出的新颖性却会受损。(55)

The reason for this is that the output of intelligent machines is essentially a statistical reproduction of human world knowledge. Large models such as ChatGPT conduct “knowledge production” based on formal logic and probabilistic inference, rather than dialectical logic and causal reasoning. Although they can efficiently screen, reorganise, and replicate human knowledge, they find it difficult to generate new knowledge [56]56 This is a citation from a source not specified in the text.. Compared with human “whimsical ideas” and “flashes of inspiration” that may transcend experience and break through thinking limitations, intelligent machines, though they can “pour out ideas” at any moment, mostly produce outputs that are “conventional and unremarkable”. In this sense, AI is a “new ordinary person”; the knowledge it produces is actually a kind of smoothed-out, mainstream knowledge acceptable to most people [57]57 “New ordinary person” (新常人) is a term coined by Chinese scholar Zhao Tingyang to describe AI as a being that reproduces average, conventional knowledge.. In work scenarios with high demands for innovation, if people are satisfied with the intelligent machine’s “good enough performance”, they often choose to “take shortcuts” or even completely let the intelligent machine “take over”. In practice, excessive dependence on AI, on the one hand, drains the interest of work that originally involved self-challenge and self-fulfilment; on the other hand, the proliferation of probabilistic, combinatorially generated “machine text” not only leads intelligent machines into “model collapse” through self-referential iteration [58]58 “Model collapse” refers to a phenomenon where AI models trained on their own generated data degrade in quality over time., but also to a large extent causes human cognitive rigidity, thinking decline, and creativity blunting. As one scholar puts it: “If we develop into an era of full-scale human–machine dialogue, our human knowledge innovation is likely to be restricted or even submerged in homogenised machine language” [59]59 This is a citation from a source not specified in the text.. If so, what Marx once pointed out—that “labour produces wisdom, but for the worker it produces stupidity and cretinism” [60]60 This is a paraphrase of Marx’s argument in the *Economic and Philosophic Manuscripts of 1844*.—is very likely to transform in the era of large-model production into a state where “AI produces mediocrity, while for the worker it produces stupidity and cretinism”.

究其缘由，智能机器的结果输出实质上是人类世界知识的统计复现。ChatGPT等大模型基于形式逻辑和概率推断、而非辩证逻辑和因果推理来进行“知识生产”,尽管其能够高效地对人类知识进行筛选、重组和复制，但却难以产生新知识。(56)相较于人类可能超出经验、突破思维限制的“奇思妙想”和“神来之笔”,智能机器虽然时刻都能“才思泉涌”,但它们的产出大多“中规中矩”。从这个意义讲，人工智能就是“新常人”,它生产出的知识其实是一种被磨平了棱角、能被大多数人接受、随大流的知识。(57)在创新性要求甚高的工作场景中，如果满足于智能机器足够棒的“劳动表现”,人类往往会选择“走捷径”,甚至完全放手让智能机器“代劳”。从实践上看，对人工智能的过度依赖，一方面会使原本自我挑战、自我实现的工作的趣味消磨殆尽；另一方面，概率化、组合式生成的“机器文本”充斥泛滥，这不仅会使智能机器因自返式迭代而走向“模型崩溃”(58),而且将在很大程度上导致人的认知固化、思维衰退和创造力钝化。正如有学者所言：“如果发展到了一个全面展开人机对话的时代，我们人类的知识创新很有可能被限制甚至淹没在同质化的机器语言中。”(59)若是如此，马克思曾指出的“劳动生产了智慧，但是给工人生产了愚钝和痴呆”(60),极有可能在大模型生产时代蜕变成为这样一种状态，即“人工智能生产了平庸，同时给工人生产了愚钝和痴呆”。

IV. Conclusion

四、结语

There is no doubt that, after entering the era of large-model production, the digital capitalist labour process is undergoing subversive changes, specifically manifested in human–machine collaboration reshaping work modes, work processes, and work content. Overall, within the capitalist system, with the support of generative AI technology, human–machine collaboration has significantly enhanced the productivity of digital capital, creating an unprecedented “digital prosperity”. But compared with digital capital’s unrestrained harvesting of the AI dividend, knowledge workers trapped in platform dependency ostensibly enjoy “re-humanised” working conditions but are in fact experiencing a more thorough “de-skilling”. Even more chillingly, even if workers one day become “cyborgs”, as long as digital capital firmly controls the key means of production in large-model production, the human–machine hybrid “superorganism” will surely become an object of exploitation by digital capital [61]61 “Cyborg” (赛博格) refers to a being with both organic and biomechatronic body parts.. In the wave of AI, hailed as “humanity’s last invention”, workers gradually reduced to “digital poor” will be the biggest losers. The root cause lies in the social life dominated by capital: “The machine is not produced as an instrument for the ‘human being’, but as an instrument for those who, through the accumulation of capital, have come to possess it” [62]62 This is a citation from Marx’s *Grundrisse*..

毋庸置疑，进入大模型生产时代之后，数字资本主义劳动过程正在发生颠覆性变化，具体表现在人机协作重塑了工作模式、工作流程和工作内容。总体上看，在资本主义制度内，在生成式人工智能技术的加持下，人机协作显著提升了数字资本的生产力，制造了史无前例的“数字繁荣”景象。但相较于数字资本恣意收割人工智能红利，身陷平台依附的知识工人表面上享有“再人性化”的劳动条件，实际上却正在遭遇更加彻底的“去技能化”。更令人不寒而栗的是，纵然有朝一日劳动者得以化身“赛博格”,但只要数字资本牢牢地掌控大模型生产的关键生产资料，人机混合的“超级有机体”定将成为数字资本榨取的对象。(61)在被称为“人类最后一项发明”的人工智能浪潮中，逐渐沦落为“数字穷人”的劳动者将成为最大输家。究其根源，在资本统治的社会生活中，“机器不是作为‘人类’的仆人，而是作为由于积累了资本从而占有机器的那些人的工具而产生出来的”(62)。

Notes:

注释：

(1) Yin Jie, “The Subjectivity Problem of Generative Artificial Intelligence”, *Social Sciences in China*, No. 8, 2024.

(1)殷杰：《生成式人工智能的主体性问题》，载《中国社会科学》2024年第8期。

(2) Accenture, *Work Models, Work Organisation, and Workers in the Era of Generative AI*, https://www.sgpjbg.com/baogao/169510.html.

(2)埃森哲：《生成式AI时代的工作模式、工作组织、工作者》，https：//www.sgpjbg.com/baogao/169510.html。

(1) Harry Braverman, *Labour and Monopoly Capital*, trans. Fang Sheng et al., Commercial Press 1978, p. 205.

(1)哈里·布雷弗曼:《劳动与垄断资本》,方生等译,商务印书馆1978年版,第205页。

(2) Wen Xiang and Zhou Xiao, ‘Western Labour Process Theory and Chinese Experience: A Critical Review’, *Social Sciences in China* 2007, No. 3.

(2)闻翔、周潇:《西方劳动过程理论与中国经验:一个批判性的述评》,载《中国社会科学》2007年第3期。

(3) Edward Ashford Lee, *Coevolution: The Future of Human-Machine Fusion*, trans. Li Yang, CITIC Press 2022, p. 335.

(3)爱德华·阿什福德·李:《协同进化:人类与机器融合的未来》,李杨译,中信出版社2022年版,第335页。

(4) Sun Weiping, ‘New Human-Machine Relations in the Digital-Intelligence Era and Their Construction’, *Journal of Hubei University (Philosophy and Social Sciences Edition)* 2023, No. 3.

(4)孙伟平:《数智时代的新型人机关系及其构建》,载《湖北大学学报(哲学社会科学版)》2023年第3期。

(5) Yin Jie, ‘The Problem of Subjectivity in Generative Artificial Intelligence’, *Social Sciences in China* 2024, No. 8.

(5)殷杰:《生成式人工智能的主体性问题》,载《中国社会科学》2024年第8期。

(6) Xiao Feng, ‘A Study of Large AI Models Based on Marx’s Theory of Objectified Labour’, *Marxist Studies* 2024, No. 4.

(6)肖峰:《基于马克思对象化劳动理论的人工智能大模型研究》,载《马克思主义研究》2024年第4期。

(7) Vivek Choudhary et al., ‘Human-AI Ensembles: When Can They Work?’, *Journal of Management*, Vol. 51, No. 2, 2025, pp. 536–569.

(7)Vivek Choudhary et al.,“Human-AI Ensembles:When Can They Work?”,Journal of Management,Vol.51,No.2,2025,pp.536-569.

(8) Mohammad Hossein Jarrahi et al., ‘Artificial Intelligence, Human Intelligence and Hybrid Intelligence Based on Mutual Augmentation’, *Big Data & Society*, Vol. 9, No. 2, 2022.

(8)Mohammad Hossein Jarrahi et al.,“Artificial Intelligence,Human Intelligence and Hybrid Intelligence Based on Mutual Augmentation”,Big Data & Society,Vol.9,No.2,2022.

(9) Paul Daugherty and James Wilson, *Human + Machine: Reimagining Work in the Age of AI*, trans. Zhao Yanan, CITIC Press 2018, p. XXVI.

(9)保罗⋅多尔蒂、詹姆斯⋅威尔逊:《机器与人——埃森哲论新人工智能》,赵亚男译,中信出版社2018年版,第XXVI页。

(10) Huawei and Tsinghua University, *AI Collaborates with Humans, Serves Humans: AI Terminal White Paper*, https://www.bgcaijingpro.com/doc/75981.html.

(10)华为、清华大学:《AI与人协作、服务于人——AI终端白皮书》,https://www.bgcaijingpro.com/doc/75981.html。

(11) He Jiang and Yan Shumin, ‘Human-Machine Division of Labour: Generative Logic, Model Types, and Mechanisms of Action’, *Nankai Business Review* 2024, No. 6.

(11)何江、闫淑敏:《人机劳动分工:生成逻辑、模式类型与作用机制》,载《南开管理评论》2024年第6期。

(12) Li Ning, ‘Unlocking AI Power: Launching a Two-Way Productive Revolution between Humans and Machines’, *Tsinghua Business Review* 2024, No. 6.

(12)李宁:《解锁AI力:开启一场人机双向奔赴的生产力革命》,载《清华管理评论》2024年第6期。

(13) Huawei and Tsinghua University, *AI Collaborates with Humans, Serves Humans: AI Terminal White Paper*, https://www.bgcaijingpro.com/doc/75981.html.

(13)华为、清华大学:《AI与人协作、服务于人——AI终端白皮书》,https://www.bgcaijingpro.com/doc/75981.html。

(14) Li Ning, ‘Unlocking AI Power: Launching a Two-Way Productive Revolution between Humans and Machines’, *Tsinghua Business Review* 2024, No. 6.

(14)李宁:《解锁AI力:开启一场人机双向奔赴的生产力革命》,载《清华管理评论》2024年第6期。

(15) *Collected Works of Marx and Engels*, Vol. 8, People’s Publishing House 2009, p. 186.

(15)《马克思恩格斯文集》第8卷,人民出版社2009年版,第186页。

(16) Fabrizio Dell’Acqua et al., ‘Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality’, HBS Working Paper, No. 24-013, 2023.

(16)Fabrizio Dell'Acqua et al.,“Navigating the Jagged Technological Frontier:Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality”,HBS Working Paper,No.24-013,2023.

(17) Sebastian Raisch and Sebastian Krakowski, ‘Artificial Intelligence and Management: The Automation-Augmentation Paradox’, *Academy of Management Review*, Vol. 46, No. 1, 2021, pp. 192–210.

(17)Sebastian Raisch and Sebastian Krakowski,“Artificial Intelligence and Management:The Automation-Augmentation Paradox”,Academy of Management Review,Vol.46,No.1,2021,pp.192-210.

(18) Paul Daugherty and James Wilson, *Human + Machine: Reimagining Work in the Age of AI*, pp. 30–31.

(18)保罗⋅多尔蒂、詹姆斯⋅威尔逊:《机器与人:埃森哲论新人工智能》,第30—31页。

(19) *Collected Works of Marx and Engels*, Vol. 8, p. 192.

(19)《马克思恩格斯文集》第8卷,第192页。

(20) *Collected Works of Marx and Engels*, Vol. 1, People’s Publishing House 2009, p. 159.

(20)《马克思恩格斯文集》第1卷,人民出版社2009年版,第159页。

(21) Ibid., p. 156.

(21)同上书,第156页。

(22) *Collected Works of Marx and Engels*, Vol. 8, p. 174.

(22)《马克思恩格斯文集》第8卷,第174页。

(23) *Collected Works of Marx and Engels*, Vol. 5, People’s Publishing House 2009, p. 417.

(23)《马克思恩格斯文集》第5卷,人民出版社2009年版,第417页。

(24) *Collected Works of Marx and Engels*, Vol. 1, p. 159.

(24)《马克思恩格斯文集》第1卷,第159页。

(25) Ibid., pp. 159–160.

(25)同上书,第159—160页。

(26) *Collected Works of Marx and Engels*, Vol. 8, p. 198.

(26)《马克思恩格斯文集》第8卷,第198页。

(27) Phil Jones, *The Post-Work Era: Labour in the Age of Platform Capitalism*, trans. Chen Guangxing, Shanghai Translation Publishing House 2023, p. 2.

(27)菲尔·琼斯:《后工作时代:平台资本主义时代的劳动力》,陈广兴译,上海译文出版社2023年版,第2页。

(28) Advait Sarkar, ‘Enough with Human-AI Collaboration’, https://arxiv.org/pdf/2306.01615.

(28)Advait Sarkar,“Enough with Human-AI Collaboration”,https://arxiv.org/pdf/2306.01615.

(29) Huang Zaisheng, ‘Generative Artificial Intelligence and New Domains of Digital Capitalist Value Movement’, *Foreign Theoretical Trends* 2024, No. 1.

(29)黄再胜:《生成式人工智能与数字资本主义价值运动新论域》,载《国外理论动态》2024年第1期。

(30) *Collected Works of Marx and Engels*, Vol. 1, p. 537.

(30)《马克思恩格斯文集》第1卷,第537页。

(31) *Collected Works of Marx and Engels*, Vol. 8, p. 199.

(31)《马克思恩格斯文集》第8卷,第199页。

(32) Ibid., p. 199.

(32)同上书,第199页。

(33) *Collected Works of Marx and Engels*, Vol. 5, p. 687.

(33)《马克思恩格斯文集》第5卷,第687页。

(34) Wei Jiang et al., ‘AI and the Extended Workday: Productivity, Contracting Efficiency, and Distribution of Rents’, NBER Working Paper, No. 33536, 2025.

(34)Wei Jiang et al.,“AI and the Extended Workday:Productivity,Contracting Efficiency,and Distribution of Rents”,NBER Working Paper,No.33536,2025.

(35) Humaid Al Naqbi et al., ‘Enhancing Work Productivity through Generative Artificial Intelligence: A Comprehensive Literature Review’, *Sustainability*, Vol. 16, No. 3, 2024, pp. 1–37.

(35)Humaid Al Naqbi et al.,“Enhancing Work Productivity through Generative Artificial Intelligence:A Comprehensive Literature Review”,Sustainability,Vol.16,No.3,2024,pp.1-37.

(36) *Collected Works of Marx and Engels*, Vol. 5, p. 743.

(36)《马克思恩格斯文集》第5卷,第743页。

(37) Ibid.

(37)同上。

(38) Ibid., p. 508.

(38)同上书,第508页。

(39) Lewis Mumford, *Technics and Civilization*, trans. Wang Keren and Li Huashan, China Architecture & Building Press 2009, p. 323.

(39)刘易斯⋅芒福德:《技术与文明》,王克仁、李华山译,中国建筑工业出版社2009年版,第323页。

(40) Wang Shuixing, ‘A Marxian Labour Theory of Value Examination of Artificial Intelligence’, *Marxist Studies* 2021, No. 5.

(40)王水兴:《人工智能的马克思劳动价值论审思》,载《马克思主义研究》2021年第5期。

(41) Jia Genliang, ‘The Third Industrial Revolution and Industrial Intelligence’, *Social Sciences in China* 2016, No. 6.

(41)贾根良:《第三次工业革命与工业智能化》,载《中国社会科学》2016年第6期。

(42) Harry Braverman, *Labour and Monopoly Capital*, p. 104.

(42)哈里⋅布雷弗曼:《劳动与垄断资本》,第104页。

(43) Larry Lohmann, “Interpretation Machines: Contradictions of ‘Artificial Intelligence’ in 21st-Century Capitalism”, in Leo Panitch and Gregory Albo (eds.), *Beyond Digital Capitalism: New Ways of Living*, London: The Merlin Press, 2021, p. 50.

(43)Larry Lohmann,“Interpretation Machines:Contradictions of ‘Artificial Intelligence’ in 21st-Century Capitalism”,in Leo Panitch and Gregory Albo (eds.),Beyond Digital Capitalism:New Ways of Living,London:The Merlin Press,2021,p.50.

(44) *Collected Works of Marx and Engels*, Vol. 5, p. 483.

(44)《马克思恩格斯文集》第5卷,第483页。

(45) Erik Brynjolfsson et al., “Generative AI at Work”, *The Quarterly Journal of Economics*, 2025, pp. 1–54.

(45)Erik Brynjolfsson et al.,“Generative AI at Work”,The Quarterly Journal of Economics,2025,pp.1-54.

(46) Ajay Agrawal et al., “Do We Want Less Automation”, *Science*, Vol. 381, No. 6654, 2023, pp. 155–158.

(46)Ajay Agrawal et al.,“Do We Want Less Automation”,Science,Vol.381,No.6654,2023,pp.155-158.

(47) Harry Braverman, *Labor and Monopoly Capital*, p. 318.

(47)哈里⋅布雷弗曼:《劳动与垄断资本》,第318页。

(48) Ethan Mollick, *Co-intelligence: Living and Working with AI*, New York: Portfolio/Penguin, 2024, p. 180.

(48)Ethan Mollick,Co-intelligence:Living and Working with AI,New York:Portfolio/Penguin,2024,p.180.

(49) Thomas Davenport and Julia Kirby, *Only Humans Need Apply: Winners and Losers in the Age of Smart Machines*, trans. Li Pan, Zhejiang People’s Publishing House, 2018, p. 10.

(49)托马斯·达文波特、茱莉娅·柯尔比:《人机共生》,李盼译,浙江人民出版社2018年版,第10页。

(50) Michael Gerlich, “AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking”, *Societies*, Vol. 15, No. 6, 2025.

(50)Michael Gerlich,“AI Tools in Society:Impacts on Cognitive Offloading and the Future of Critical Thinking”,Societies,Vol.15,No.6,2025.

(51) Yu Xue, “Automatic or Autonomous: On ‘Automated Survival’ in the Age of Artificial Intelligence”, *Exploration and Free Views*, No. 2, 2024.

(51)于雪:《自动还是自主——论人工数智时代的“自动化生存”》,载《探索与争鸣》2024年第2期。

(52) Eva A. M. van Dis et al., “ChatGPT: Five Priorities for Research”, *Nature*, Vol. 614, 2023, pp. 224–226.

(52)Eva A.M.van Dis et al.,“ChatGPT:Five Priorities for Research”,Nature,Vol.614,2023,pp.224-226.

(53) Nan Jia et al., “When and How Artificial Intelligence Augments Employee Creativity”, *Academy of Management Journal*, Vol. 67, No. 1, 2024, pp. 5–32.

(53)Nan Jia et al.,“When and How Artificial Intelligence Augments Employee Creativity”,Academy of Management Journal,Vol.67,No.1,2024,pp.5-32.

(54) Ethan Mollick, *Co-intelligence: Living and Working with AI*, p. 119.

(54)Ethan Mollick,Co-intelligence:Living and Working with AI,p.119.

(55) Anil R. Doshi and Oliver P. Hauser, “Generative AI Enhances Individual Creativity but Reduces the Collective Diversity of Novel Content”, *Science Advances*, Vol. 10, No. 28, 2024, pp. 1–9.

(55)Anil R.Doshi and Oliver P.Hauser,“Generative AI Enhances Individual Creativity but Reduces the Collective Diversity of Novel Content”,Science Advances,Vol.10,No.28,2024,pp.1-9.

(56) Guglielmo Carchedi, “Beating the Untrodden Paths: Computers, Artificial Intelligence and Quanta in Marxist Theory”, *Historical Materialism*, Vol. 32, No. 2, 2024, pp. 323–353.

(56)Guglielmo Carchedi,“Beating the Untrodden Paths:Computers,Artificial Intelligence and Quanta in Marxist Theory”,Historical Materialism,Vol.32,No.2,2024,pp.323-353.

(57) Liu Hailong, “Generative AI and Knowledge Production”, *Editorial Friend*, No. 3, 2024.

(57)刘海龙:《生成式AI与知识生产》,载《编辑之友》2024年第3期。

(58) Ilia Shumailov et al., “AI Models Collapse When Trained on Recursively Generated Data”, *Nature*, Vol. 631, 2024, pp. 755–759.

(58)Ilia Shumailov et al.,“AI models Collapse When Trained on Recursively Generated Data”,Nature,Vol.631,2024,pp.755-759.

(59) Tao Feng and Liu Xingchen, “From Human-Machine Dialogue to Human-Machine Interaction: A Philosophical Reflection on Large Language Models of Artificial Intelligence”, *Social Science Front*, No. 5, 2024.

(59)陶锋、刘星辰:《从人机对话到人机交往——人工智能大语言模型的哲学反思》,载《社会科学战线》2024年第5期。

(60) *Collected Works of Marx and Engels*, Vol. 1, p. 159.

(60)《马克思恩格斯文集》第1卷,第159页。

(61) Michael E. Gardiner, “Critique of Accelerationism”, *Theory, Culture & Society*, Vol. 34, No. 1, 2017, pp. 29–52.

(61)Michael E.Gardiner,“Critique of Accelerationism”,Theory,Culture & Society,Vol.34,No.1,2017,pp.29-52.

(62) Harry Braverman, *Labor and Monopoly Capital*, p. 172.

(62)哈里⋅布雷弗曼:《劳动与垄断资本》,第172页。

Online editor: Zhang Jian

网络编辑：张剑