Huang Zaisheng: Large Model Production and the Contemporary Interpretation of Marx's Labor Theory of Value
Marxist labor theory of value holds that living labor is the sole source of value creation. Since the establishment of the capitalist mode of production, capital has spared no effort in utilizing technological innovation and organizational change to achieve the discipline and absorption of labor. Meanwhile, alongside the rise in the organic composition of capital, the machine system—acting as objectified labor—dominates living labor. Consequently, the substantiality of value is increasingly rendered hollow within ever-more automated production. Overall, capital strives to acquire new paths and methods to devour living labor through the "spatio-temporal fix" [1] of world market expansion and the broadening of production sites, all to alleviate the crises of capitalist value movement and sustain the capitalist social production system.
With the accelerated iteration and everyday application of General Artificial Intelligence (AGI) represented by ChatGPT, digital capitalism has entered the era of Large Models. Compared to traditional large-scale industrial production, "Large Model production" refers to a mode driven by AGI technology, carried out via platform production networks, and supported by the accelerated formation and dialectical movement of "new quality productive forces" and new relations of production. It drives the high-level automation of both material and spiritual production under capitalism, thereby efficiently achieving the hyper-personalized customization and "one-click" production of goods and services. Under Large Model production, the agency of digital machines becomes prominent, intuitively dissolving human irreplaceability and driving complex changes in the man-machine relationship of digital capitalism. As some scholars have noted, the "human-like intelligence" and autonomy of intelligent systems are steadily growing, and they are acquiring increasingly diverse and powerful "labor capacities." Superficially, the "subjectivization" of digital machines and the "departure" of human labor increasingly create a digital production landscape of "full automation," "zero labor," and "post-work." Naturally, the argument that Marxist labor theory of value is obsolete or invalid has resurfaced, causing the theory—which centers on the question of "what creates value"—to encounter unprecedented challenges. Among these, the rhetoric that views digital machines evolving into "AI Agents" as a "new species"—thereby redefining "labor" and "laborer" and overturning the stipulations of subjectivity—is theoretically deceptive.
In view of this, to promote the "upholding of fundamentals and breaking of new ground" in Marxist labor theory of value, it is urgent to adhere to the core essence that "labor is the substance and immanent measure of value" while scientifically understanding the nature of digital machines in the era of Large Models. We must acutely grasp the deformation of digital labor, deeply analyze the internal tensions and future prospects of the value movement in digital capitalism, and strive to deepen and solidify the contemporary interpretation of labor creating value.
I. Large Model Production and the Triple Paradox of Value Movement
Practice shows that the dialectical movement of the relationship between value and "anti-value" [2] within capitalist social production consistently causes the inherent contradictions of capital's self-valorization to play out and intensify. From the perspective of the substance of value, objectified labor increasingly dominates living labor, leading to a "production paradox" of expanding wealth and shrinking value. From the perspective of value realization, the movement of capital creates a general devaluation of labor power, leading to a "market paradox" of expanding production and shrinking demand. From the perspective of the existence of value, capitalist production becomes increasingly socialized, leading to a "systemic paradox" where common elements emerge while exchange relations persist. Following the implementation of Large Model production, the trends of intelligent production, labor precarity, and platform infrastructuralization have further deepened, making the internal tensions of the value movement in digital capitalism increasingly difficult to reconcile.
1. Large Model production amplifies the "production paradox" of value movement
With the vigorous development of large-scale mechanized industry, an inevitable trend of capitalist value movement is that "objectified labor, in the labor process itself, confronts living labor as the force that dominates it"; meanwhile, living labor is transformed into a "mere living appendage" of the machine system. Thus, from the perspective of the production process of value formation, capital inevitably leads to the "hollowing out" of the substance of value in its pursuit of self-valorization. Figuratively speaking, this is as if the capitalist value movement contains a "black hole" that only devours old value but cannot spit out new value. As the American scholar David Harvey has pointed out, if living labor is the source of value and profit, then replacing living labor with dead or mechanical labor makes no sense politically or economically. In Marx's view, this is a core contradiction of capitalism.
Reviewing the history of capitalist production, under Fordism, the capitalist application of science and technology continuously promoted standardization and automation, achieving a fundamental reversal of the man-machine relationship: "In manufacture and handicrafts, the worker uses a tool; in the factory, the worker serves the machine." Yet, labor as a "conscious linkage" remained indispensable. With the development of capitalism, immaterial labor unfolded within people's daily communication, social cooperation, and emotional exchange; for a time, capital found it difficult to achieve the discipline and absorption of such labor, let alone use an "active machine system" to replace it. Entering the stage of digital capitalism, intelligent production stepped into the fast lane, greatly expanding the operational boundaries of "replacing humans with machines," and direct labor in the production process was rapidly reduced. However, digital economic practice shows that driven by analytical AI technology, intelligence-enhanced digital machines are not only confined to specific operational fields, but the increasingly exposed "ghost work" [3] serves as a constant reminder: the so-called "unmanned," "zero labor," and "post-work" scenarios remain merely a digital utopian imagination of technological optimists.
In fact, in the Marxian context, even if capital obtains the most perfect and suitable machine system—namely, an automated machine system—it does not mean the total withdrawal of labor, but rather its distribution as a link "at many points of the mechanical system." Moreover, in Marx's view, "a capitalist who used no variable capital at all within his own branch of production, and therefore no workers at all" was merely an extreme hypothesis. As mentioned above, along with the transition of capitalist production from mechanization to automation and then to intelligentization, the form of labor has also evolved from material labor to immaterial labor and then to digital labor; the living labor that creates value has never truly departed. "The steam engine, the spinning machine, electrification, and electronization have never continuously reduced the amount of labor; on the contrary, they have increased it." But all things change with the times. After the implementation of Large Model production, bolstered by AGI technology, the decision-making autonomy and capability generalization of digital machines are improving at a pace beyond ordinary imagination. Setting aside highly controversial topics such as "superintelligence," "machine consciousness," and "Life 3.0," from the perspective of the capitalist production process, today's digital machines demonstrate a "high-level automation" capability for continuous iteration, causing more and more cognitive labor to be replaced by hyper-efficient intelligent algorithms. It can be said that "previous industrial automation systems or intelligent machinery mainly liberated human physical strength and replaced manual workers, but AGI like ChatGPT is disrupting human mental labor and replacing various tasks undertaken by 'knowledge workers'."
Facing the immense power of the machine system in the industrial age, Marx once exclaimed: "If the machine were to annihilate the whole class of wage-workers, how terrible that would be for capital, which, without wage-labor, ceases to be capital!" Indisputably, the productive might displayed by digital machines in the Large Model era dwarfs all industrial machinery to date. With the technological refinement and scenario application of AGI, "Capital is now reshaping the fundamental capacities of 'living labor'—cognition and perception—into machine forms suitable for capital; it is dedicated to intensifying the capture of these basic capacities and accelerating their transformation into 'distributed' intelligent systems." If this trend continues, an unavoidable theoretical question is: how many opportunities will be left for ordinary workers even to "stand beside the production process" and tend the machines?
In response, arguments such as the "useless class" are quick to conclude the future fate of laborers, but there is little discussion regarding what the situation will be for capital. In the author's view, as digital capitalism's AGI technology accelerates and the "total replacement of living labor by dead labor" pushes toward the limit, the approach of the capitalist value movement toward an "explosive prospect" will be irreversible. Consequently, once capital movement under Large Model production drifts further away from labor, "production based on exchange value" will fall into production disorder and market turmoil until it reaches economic collapse.
2. Large Model production amplifies the "market paradox" of value movement
In the context of political economy, value reflects the social relationship through which people exchange labor under conditions of a commodity economy. It originates from the opposition between private labor and social labor under the social division of labor and takes shape through the unity of commodity production and commodity circulation. In other words, as a form of social validation, value achieves the dual transformation of concrete labor into abstract labor and private labor into social labor through the mediation of exchange value. Furthermore, the character of "labor as the objectification of general labor and as a [means of] satisfying general needs" is affirmed through exchange, which naturally becomes a life-or-death matter for sustaining the capitalist value system.
From the perspective of value realization, with the rapid development of capitalist productive forces and the increasingly abundant supply of commodities, market development and demand expansion become more urgent. However, hundreds of years of capitalist economic development show that although capital uses any means possible to open domestic and international markets, "relative overproduction" always accompanies socialized large-scale production. At its root, the movement of capital itself contains an insurmountable "market paradox": on the one hand, to acquire value, capital strives to create an increasingly rich supply of commodities; on the other hand, to accelerate self-valorization, capital does its utmost to devalue labor power. As a result, the expansion of capitalist production and the shrinkage of demand move in opposite directions, causing interruptions in the movement of value and triggering periodic crises.
This is concentrated in the following: under Fordism, the standardization and streamlining of production continuously promoted the "deskilling" of labor, where "the worker becomes a mere appendage of the machine, and it is only the simplest, most monotonous, and most easily acquired knack that is required of him." Thus, "owing to this shift, the worker's own labor capacity is devalued." After the rise of Post-Fordism, capital pushed for neoliberal institutional innovations, continuously promoting employment precarity, such that more and more ordinary laborers could only scrape by on consumer credit. Entering the stage of digital capitalism, the rise of digital labor brought new employment methods and opportunities, but the platform-based organization and algorithmic management of digital labor further intensified employment instability.
As AGI technology begins to popularize, production processes across all industries in digital capitalism are undergoing reorganization and reshaping. McKinsey estimates that in 63 use cases, generative AI applications could bring between $2.6 trillion and $4.4 trillion in potential value to the global economy annually. Currently, digital tech giants—through comprehensive monopolies on data, computing power, and algorithms—firmly control the techno-economic ecosystem of AGI, attempting to monopolize the new round of digital dividends released by Large Model production. It is certain that if the current platform system of digital capitalism does not change course, more and more workers will inevitably fall into the living predicament of the "digital poor" during the digital transformation.
This is because, first, under Large Model production, the phenomenon of "replacing humans with machines" continues to spread to the field of capitalist knowledge production; the "creative class" possessing intellectual expertise will follow in the footsteps of blue-collar workers, abandoned by a ruthless yet narratively adept digital capital. Second, AGI technology further activates "Digital Taylorism" [4]; digital capital accelerates the dismantling of the human cognitive labor process, continuously standardizing, modularizing, and automating computable work content, thereby "deskilling" even high-level capitalist labor. It can be expected that the simplification of complex labor in digital capitalism will accelerate, and the continued decline of labor's share in national income distribution will be an inevitable trend.
As a result, on the one hand, Large Model production realizes the industrialized production of knowledge commodities, and combined with AGI technology empowering traditional manufacturing to improve quality and efficiency, the commodity and service capacity of digital capitalism will show exponential growth. On the other hand, Large Model production will squeeze an enormous number of knowledge workers out of the direct production process of capitalism, causing the "middle class"—which has long supported mass consumption in developed capitalist countries—to collapse at an accelerated rate. Consequently, Large Model production further amplifies the "market paradox" of expanding capitalist production and contracting demand; digitalized capital production and circulation are destined to encounter more frequent overproduction and more violent market fluctuations.
3. Large Model production amplifies the "systemic paradox" of value movement
Under the capitalist mode of production, the products of labor universally become commodities, and value regulates social production as a dominant and governing force. Viewed through the practical logic of the movement of value, capital creates an antithesis between private labor and social labor by monopolizing the means of production, thereby allowing value—as a social reality—to manifest within the "common element" of market exchange. Simultaneously, however, capital continually imparts a scientific character to production; by driving the historical transformation of the instruments of labor, it acquires immense power, objectively promoting the increasing socialization of capitalist production. Consequently, the more capitalist "objectified productive forces" develop, the more the boundaries between private and social labor blur. The production of use-values that are direct and do not enter exchange gathers momentum, and the limits of production based on capital become increasingly apparent.
Substaintially, the "institutional paradox" [5] within the capitalist movement of value is the concrete manifestation of the contradiction between the private ownership of the means of production and large-scale socialized production. Under Fordism, continuously improving systems of industrial machinery absorbed general social knowledge, and "the productive power of capital develops with this general progress, which capital appropriates free of charge." At the same time, the increasing concentration of production became incompatible with the form of capital's command over production. Entering the era of cognitive capitalism, the "living knowledge" found in daily interaction became a new "profit element" that the capitalist movement of value was eager to devour. Capital pushed for the commodification of knowledge production and the creation of intellectual property regimes, seizing general intellect through accumulation by dispossession. Yet, the narrow private interests of capital greatly constrained the full release of social productive forces. Moving into the stage of digital capitalism, practices of collaborative network production based on the commons—such as Wikipedia and Free and Open Source Software (FOSS)—demonstrate the realistic possibility of human social production transcending the logic of capital. Digital capital utilizes platform power to carry out massive "digital enclosures," obsessively pursuing multilateral market monopolies and transforming digital cyberspace—which harbors immense potential for social cooperation and co-creation—into a new field of social production subordinate to capital. As a result, "platform capitalism in Silicon Valley has not only dramatically changed the overall meaning of the sharing economy, turning groups of formerly happily cooperating peer volunteers into a frustrated army of unpaid or low-paid laborers, who now must deal with the catastrophic consequences of an unexpected discovery: the commodification and exploitation of the commons."
With the normalization of large-model production, a prominent feature of digital capitalist social production is that large models, as new means of production, are becoming further integrated into platform infrastructure. They are becoming foundational elements of social production, much like water and electricity in the industrial age or software infrastructure in the internet era. Furthermore, supported by General Artificial Intelligence (GAI), digital machines are reaching maturity, and the social productive force bursting from the general intellect—produced as the "immediate organ of the real life process"—is realized more efficiently. Thus, large-model production drives the means of production toward increasing infrastructuralization; both material and mental production in digital capitalism are more socialized than ever before. From the perspective of the base of the mode of production, the inherent tension in digital capitalism—where "the production of use-value is limited by exchange-value"—becomes increasingly manifest and acute. As Phil Jones pointed out, "Today, the viability of the capitalist system is questionable not only politically but ontologically."
Theoretically, developing the digital commons with the aid of large-model production could promote open data, shared computing power, and open-source algorithms, thereby releasing intelligent productive forces to the maximum extent to better benefit human society. In reality, however, the truth about AI is that digital tech giants are obsessed with building digital empires, keen on AI "arms races," and actively opening new commercial territories. Consequently, while digital capital exerts every effort to maintain the "social form of use of the material means of production" suited to its own valorization, it also prunes, edits, and suppresses the institutional genes of co-creation and sharing inherent in large-model production, preventing them from breaking through. The limits of capitalist production aimed at value thus become more exposed to the world.
II. The Debate over General Agents and the Subjectivity of Digital Machines
In the industrial age, the capitalist machinery system displayed immense power to create wealth; "it seemed that the result was to make material forces gifted with intellectual life, while human life was reduced to a stultifying material force." Yet facts proved that within the movement of value in industrial capitalism, the human was always the subject; increasingly automated industrial machinery achieved almost nothing in substituting for complex human labor. In the stage of digital capitalism, supported by the technology of analytical AI, digital machines began to demonstrate stunning performances in specific fields that completely surpassed humans. After the sudden emergence of generative AI, represented by ChatGPT, digital machines simulated human mental production and presented a trend toward subjectivity that could thoroughly replace human labor. Consequently, in the eyes of some, faced with increasingly autonomous digital machine systems, the reconstruction of Marx’s labor theory of value appears imminent.
1. The Leap of Digital Machines into General Agents
Marx pointed out that the machinery system of industrial capitalism consists primarily of the motor, the transmitting mechanism, and the working machine. "When the working machine performs all the movements necessary for processing the raw material without human assistance, and only requires oversight from the worker, we have an automatic system of machinery." Throughout the history of capitalist machinery, regardless of the degree of autonomy industrial machines displayed, their essence as instruments of labor suited to the requirements of capital and subordinate to the "human world" remained unchanged. In the stage of digital capitalism, with the iterations and accelerated application of AI technology, algorithm-driven control mechanisms have been integrated into machine apparatuses, and the autonomy and automaticity of digital machines have grown daily. In particular, with the rapid development of large-model technology, the autonomous ability of digital machine systems to perceive environments, form memories, make decisions, and take action is rising fast; the leap toward becoming General Agents has become unstoppable.
From current developments, a General Agent is a complex integration of hardware and software, primarily composed of four key components: a Large Language Model (LLM), memory, task planning, and tool use. Among these, the LLM is the core brain-like device that directly determines the capacity radius of the General Agent. The memory component serves as a storage device, providing support for the agent to develop abilities such as understanding, association, and reasoning through deep learning and reinforcement learning. The task planning component ensures that the General Agent remains goal-oriented during situational interactions, while the tool-use component helps it complete tasks efficiently in a "human-like" manner. It can be expected that in the future, as GAI technology is optimized and upgraded, General Agents will not only become more autonomous and adaptable but their situational interactions with other agents or humans may also give rise to a form of "sociality."
In view of this, the appearance and leap of General Agents further promote the objectification of the subject and the subjectification of the object. From the surface of digital capitalist production, as large-model production continues to develop, all human labor (not just direct or individual labor) sees its "power to create value tend toward disappearance as an infinitesimal quantity," yet digital capital remains immensely successful in commercial gold-digging. Theoretically, "what creates value" has become a topic requiring serious discussion. As mentioned above, once digital machines leap into General Agents and participate in the labor process in a more "human-like" way, they will inevitably amplify the inherent contradictions of the capitalist movement of value. Clarifying the essence of digital machines represented by General Agents and explaining their role in value creation is particularly urgent and important for upholding the fundamentals and breaking new ground in Marx’s labor theory of value.
2. Three Responses to Whether General Agents Create Value
Fundamentally, the foundation of Marx’s labor theory of value lies in the irreplacability of human labor. At present, digital machine systems are leaping toward becoming General Agents, beginning to show the productive temperament of a "perfect laborer." Within the discourse of Marx’s labor theory of value, "admitting that AI can completely replace human labor means the negation of Marx’s labor theory of value." Facing this theoretical dilemma, the approach of arbitrarily claiming that Marx’s labor theory of value is obsolete or invalid and starting anew is not worth refuting. Conversely, ignoring new circumstances and remaining satisfied with "textbook" interpretations of principles is merely self-stultification. By contrast, a preferable research path is to inherit Marx's discursive system, attend to the new changes in the movement of value under digital capitalism, and attempt a theoretical reconstruction of the labor theory of value. Broadly speaking, three representative viewpoints regarding the "question of the era"—whether General Agents create value—merit detailed analysis.
The first is the "Interpretive Machine" theory. The British New Left scholar Larry Lohmann has pointed out that, under capitalist application, machine systems in the digital age are essentially no different from those in the industrial age. In the industrial age, capital pushed for the application of science to technology, forming objectified intellectual power—the industrial machine—to dismantle, imitate, and replace manual skills in the production process. In the digital age, capital pushes for the application of AI to knowledge production, forming objectified algorithmic power—the interpretive machine—to decompose, reorganize, and replicate human interpretive abilities. Furthermore, "artificial intelligence" in the 21st century is essentially the digital mechanization of human interpretive capacity. From the context of the labor theory of value, the key lies in the fact that although the "digital mechanization" of capitalist production today is faster, wider in scope, and more covert in its diffusion, the "interpretive machine" has not made living labor "obsolete" compared to the industrial era. Since the machine systems of digital capitalism have not undergone a qualitative change and remain essentially objectified intellectual power subordinate to capital, Marx’s view of machinery remains valid, and the so-called "reconstruction" of the labor theory of value is much ado about nothing.
In the author's view, the "interpretive machine" theory correctly identifies that the technical reality of AI’s capitalist application has not transcended the domain of Marx’s labor theory of value. Overall, however, this view holds a linear mindset regarding technical progress. It fails to fully recognize that the new generation of information technology, especially GAI, is advancing at an exponential pace and bringing revolutionary, disruptive shocks to the capitalist production process. Consequently, its theoretical identification of the technical essence and productive status of digital machines in the world today is somewhat too conservative. For this reason, facing the dizzying iterations of digital machines in the era of large models, it is difficult to use the logic of the "interpretive machine" theory to provide a convincing theoretical response to the triple paradox of the movement of value in digital capitalism.
The second is the "New Laborer" theory. When discussing the difference between machines and tools in capitalist production, Marx wrote: "The machine, which possesses skill and strength in place of the worker, is itself the virtuoso, with a soul of its own in the mechanical laws acting through it; and it consumes coal, oil etc. (materials of consumption), just as the worker consumes food, to keep up its continuous motion." Undoubtedly, this anthropomorphic description of industrial machines inspired endless imagination regarding the potential development of capitalist machinery systems. Entering the stage of digital capitalism, AI technology and its applications change daily. It can be said that digital machines are increasingly possessing "human-like intelligence" through the statistical laws acting within them.
In view of this, an increasing number of scholars argue that examining the human-machine relationship in digital capitalism requires breaking the myth of "anthropocentrism," discarding the perception of AI as a pure tool, and seriously considering the "subjectivity" of intelligent machines by redefining "labor" and "laborer." In this regard, some scholars believe that even in the stage of Weak AI, the capitalist production process was already a new form of labor, and Weak AI was a new type of laborer. Generative AI, represented by ChatGPT, should be defined as a "New Person" (xin changren). From the perspective of value theory, if a General Agent can handle all human work and "intelligent systems possess a certain 'species advantage' over natural persons," then it is not surprising that General Agents can also engage in "labor." Furthermore, if certain socio-historical conditions exist to drive the "proletarianization" of AI and the requirement that it "must possess its own necessities of life," then the question of whether "the working machine brain is possessed as a means of production (slave) or hired as free labor" evolves toward a relation of production suited to capital, and it becomes logical for General Agents to create "value."
In summary, the basic logic of the "new worker" theory is as follows: so long as it is accepted that general agents are employed by digital capital and become "producers" and "consumers" in the production of Large Models, the "production paradox" and "market paradox" within the movement of value will not be infinitely amplified to the point of spiraling out of control. Accordingly, the digital capitalist value system could regain its vitality through the addition of "new labor elements." Yet, it is not without irony that, if pushed according to this view, capital’s promotion of the objectified accumulation of social labor and general intellect would instead allow capitalism to function just as well without human participation, eventually evolving into a perpetual capitalism based on artificial intelligence. In the author's view, the greatest flaw of the "new worker" theory lies in its rejection of the human nature [13] of historical development. It crudely transplants socio-historical categories such as "labor" and "value" into its imagined "new world" of digital capitalism, only to end up being too smart for its own good: by "introducing artificial intelligence machines into categories representing relations between people," it renders the concept of capital unrecognizable. Marx once pointedly noted: "man no longer does that which he can have things do for him, — once that point is reached, the historical mission of capital is completed." Undoubtedly, this maxim of the century has long been forgotten by the adherents of the "new worker" theory.
The third theory is the "refusal of labor." The Marxian view of labor points out that "labor created man himself." Fundamentally, the purposiveness and autonomy of human labor distinguish human practical activity from the instinctive activity of animals. Humans in reality are the sole subjects of labor. For a long time, although the capitalist application of machine systems has continuously crowded out direct labor, human labor has demonstrated a unique species-essence in terms of creativity, criticality, and reflexivity, such that the core elements of living labor cannot be replicated, reorganized, or replaced by objectified intellectual forces. Precisely for this reason, the status of the human being as the subject of production has not been fundamentally shaken.
However, times have changed. As mentioned above, with the technical empowerment of General Artificial Intelligence, digital machines have already demonstrated an astonishing "human-like intelligence." Faced with the intelligent evolution of digital machines and the resulting "crisis of subjectivity" for humans, a theoretical question that is far from absurd arises: if one day all qualities of labor in a positive sense can be replicated, imitated, or even surpassed by highly evolved machine systems, and consequently all work that humans are capable of can be handed over to digital machines, can the claim that "all value comes from human labor" still hold true?
In the view of American scholar George Caffentzis, in the era of artificial intelligence, continuing to interpret Marx’s labor theory of value through the positive aspects of human labor will only lead into a theoretical embarrassment of self-inconsistency. That is to say, if one follows the basic viewpoints of Marx's labor theory of value and applies a bit of theoretical deduction, one arrives at the absurd yet logical conclusion that "machines can also labor and create value." In view of this, to promote an interpretation of Marx's labor theory of value that advances with the times, it is necessary to shift research perspectives and establish new grounds for argument. He argues that from the negative aspect of human labor, the laborer’s ability to "refuse labor" is precisely what neither animals nor machine systems possess. Furthermore, within the domain of Marxian value theory, the reason only labor creates value is not because human labor cannot be completely replaced by machines, but because the ability to "refuse labor" ensures that human subjectivity can never be usurped by the capitalist machine system.
The "refusal of labor" theory adheres to its understanding of the basic propositions of Marx’s labor theory of value and, by attending to new circumstances and responding to new challenges in the digital age, attempts to find "theoretical blind spots"—this is indeed commendable. However, judging from its basis of argument, attempting to prove that Marx's labor theory of value remains valid based solely on the so-called "new discovery" of the "refusal of labor," while abstracting away from specific social relations, is merely a subjective assertion that disregards the socio-historical determination of the movement of value. It ends up falling into the embrace of subjective value theory. To take a step back, what is even more critical is that under the production of Large Models, once ordinary laborers have almost no market opportunities left even to sell their own labor-power, how can the ability to "refuse labor" be of any use?
3. General agents are essentially production proxies
Marx pointed out that under the capitalist mode of production, industrial machinery is essentially objectified intellectual power appearing in the form of fixed capital. Entering the stage of digital capitalism, the autonomy and automaticity increasingly displayed by digital machines, represented by general agents, bring an unprecedented practical impact on the status of the human as the subject of production. Whether the general agent is still a "technical being" or a "new species" capable of labor has become the primary question that must be answered to uphold Marx's labor theory of value. Theoretically, as long as the understanding of the "human-likeness" and "machine-likeness" of general agents remains ambiguous or even wavering in its stance, promoting an interpretation of Marx's labor theory of value that advances with the times will yield no substantial new progress.
In Marx's context, the determination of human subjectivity is manifested in two aspects: first, as a subject of the cogito; second, as a subject possessing sociality and practical nature. From the path of technical evolution, as long as the continuous evolution of general agents does not break away from the "Turing machine" paradigm, then "artificial intelligence is merely a product that abstracts a certain aspect of human intelligence and develops and inflates it one-sidedly." In a statistical sense, the "human-like intelligence" of general agents is actually a "probability prediction model + stochastic optimization." The core dimensions of human intelligence, such as non-verbal intelligence, emotional decision-making, innovation, and intuition, are still mastered by humans. This determines that "within the domain of Marx's labor theory of value, the essence of the human person dictates that artificial intelligence can never possess the dual subjective status of a human social producer and consumer." Furthermore, no matter how digital machines represented by general agents develop, they remain means of labor serving capital and are subject to the "world of man." In other words, general agents remain "a component of constant capital," a tool of production, rather than becoming an independent source of value. Of course, considering that compared to the machine systems of the industrial era, the machine systems of the digital era participate in the labor process in a more "human-like" way, general agents are essentially just "production proxies under human supervision."
III. AI Augmentation and the Value Creation of Digital Labor
In recent years, with the deep development of artificial intelligence technology, more and more human labor has realized a shift from "embodiment" to "disembodiment" [14]. However, the development of the platform economy shows that while the "unmanned" production of digital capitalism is advancing triumphantly, it has not "completely abolished human labor"; "hidden behind the algorithms, thought to be automated, they in fact remain highly dependent on human labor." Furthermore, digital labor has become a new source of value creation in the stage of digital capitalism. The arrival of the era of Large Models has accelerated the democratization of AI technology, and the man-machine relationship in digital capitalism is undergoing complex changes. With the help of the "inorganic body" of the human being—the general agent—the value creation of digital labor also presents new changes and characteristics that require urgent attention.
1. The increasing importance of unpaid user labor
Since Web 2.0, the platform revolution has accelerated the upgrading of the digital existence of the general public. Under the digital capitalist platform system, unpaid user labor, as a typical form of digital labor, has become an important method for digital capital to discipline and subsume labor. Broadly speaking, hundreds of millions of active users linger in the colorful virtual world of the network, but inadvertently become the "network digital slaves" of digital capital, playing an irreplaceable role in data element supply, digital space production, digital machine manufacturing, and the production and consumption of digital commodities. From the perspective of Marx’s view of machinery, capital has always tirelessly pursued the free appropriation of general intellect; in the digital age, through the objectification of unpaid user labor, this is realized in a more real-time and convenient manner.
Following the promotion of Large Model production, more and more AI-native applications are joining the platform economy ecosystem. In practice, natural language conversational human-computer interaction has ignited "AI's iPhone moment," disruptively reshaping user operation interfaces. This not only brings a brand-new digital consumption experience to users but also makes powerful intelligent tools readily available. Unlike the tools of mediated labor in Marx's context, the general agent as a production proxy is itself a "skilled craftsman"—a near "omnipotent" digital assistant that efficiently "works for you and solves your problems" in knowledge-intensive jobs. As a result, the "Dialogue as a Service" spawned by Large Model production has greatly lowered the threshold for the general public to access the network; the digital existence of the masses is leaping toward an intelligent existence, and the value contribution of unpaid user labor will become increasingly prominent. This is concentrated in the following two aspects. First, high-quality corpora are the basic element conditions for the generation of Large Model intelligent productive forces. Fundamentally, the "human-like intelligence" of general agents is a statistical expression of the general intellect of human society. Moreover, the "intellectual evolution" of general agents requires continuous data feeding to capture in real-time the new accumulations following the development of human civilization. Currently, the development of Large Models dominated by digital technology giants has almost exhausted all of humanity's information and knowledge reserves. Thus, from the perspective of the sustainable supply of data elements, the massive amount of data being generated every moment by active users in their digital existence carries the increments of the general intellect of human society, and will naturally be increasingly relied upon for the development and iteration of Large Models. In particular, with the help of general agents, active users can cross the limitations of "technique" and "efficiency," focusing on the creative conception and sincere expression of digital performance, realizing the rapid transformation from "inspiration—clue—work," and the quality and richness of User Generated Content (UGC) will also continuously improve. Compared with the "machine text" generated by AIGC, the data value of UGC in Large Model production is even more anticipated. Second, from the perspective of digital machine manufacturing, the production and evolution of general agents are realized by using industry-proprietary data to fine-tune the foundation of pre-trained Large Models to form proprietary models. In practice, to ensure the safety and reliability of Large Model production results, the parameter modeling, feedback learning, and value alignment of general agents are inseparable from manual guidance and supervisory intervention. Within this process, in the move toward intelligent existence, every close contact between the user and the general agent—whether chatting and asking questions or providing feedback and correcting errors—is, without exception, contributing bit by bit [15] to the upgrading and iteration of digital machines without compensation.
2. The narrowing of online crowdsourced labor
Entering the stage of digital capitalism, a typical form of labor in the global digital labor landscape is online crowdsourced labor, also known as "cloud labor." Based on the degree of professional skill required to complete digital tasks, crowdsourced labor is further subdivided into freelancers, represented by Upwork, and micro-labor, represented by Amazon Mechanical Turk. Fundamentally, crowdsourced laborers still engage in various forms of cognitive labor; the labor process is simply "de-spatialized and de-temporalized" under the platform system. Since the implementation of Large Model production, the substitution of labor by digital capital has extended into the field of human knowledge production. Currently, the artificial intelligence generated content (AIGC) capabilities possessed by general agents are accelerating their iteration and evolution toward multi-modality, broad fields, and high coordination. Consequently, the digital tasks and online work content released by digital capital will undergo fission and reorganization, thereby causing the value contribution and the fate of the producers among crowdsourced laborers to experience variables.
On the one hand, for crowdsourced micro-workers engaged in "ghost work," low-end cognitive labor such as text generation, intelligent customer service, and data labeling will, in time, eventually be replaced by the ultra-efficient and more economical general agents. Furthermore, digital capital—shrouded in the halo of high technology—will impatiently seize the new opportunities brought by Large Model production to distance itself as much as possible from entanglements with digital labor, so as to more complacently broadcast the "alchemy of AI" everywhere. Of course, in the short term, general agents will not truly achieve understanding, thinking, and reasoning like humans; crowdsourced micro-labor, which specializes in compensating for the capacity shortcomings of digital machines, will not disappear rapidly. However, overall, it will cluster in special production or service areas, presenting a new trend of a niche labor population and specialized work requirements. This is concentrated in the following: while traditional micro-work opportunities are accelerating their decline, digital task demands such as data labeling for niche cultural scenes and professional industry fields, as well as the establishment of learning templates in Large Model development, will all spawn new online jobs. Thus, within the profit chain constructed by digital capital, the production contribution of crowdsourced labor—accelerating its shift from labor-intensive to knowledge-intensive—will increase, but simultaneously, online work opportunities for globally vulnerable employment groups will significantly shrink.
On the other hand, for freelancers relying on digital platforms, large models have derived a rich matrix of capabilities, assisting them in extracting themselves from routine digital technical maneuvers. This affords them more time and energy to engage in more imaginative and challenging digital labor. At the same time, the massive amount of knowledge and content generated by general agents through prompting and induction often contains insights or creative inspirations that open up new horizons of thought. Consequently, under large model production, possessing general technical expertise is no longer a competitive weapon for freelancers; rather, "makers" who are good at "manufacturing ideas" and proficient in using intelligent tools will be in high demand during the Fourth Industrial Revolution. Practice shows that super-individual owners who fully utilize AIGC and Robotic Process Automation (RPA) technologies have already begun to emerge.
3. A new turning point for the decencies of gig labor Entering the stage of digital capitalism, the gig economy has experienced explosive growth, giving birth to app-based gig labor. The number of app-based gig workers, such as ride-hailing drivers, food delivery riders, and couriers, has grown sharply. Driven by platform algorithmic systems, they rush through the streets and alleys of cities regardless of rain or shine, becoming an indispensable part of people's digital existence. Fieldwork indicates that gig workers, as members of the "precariat" [16], immerse themselves fully in the platform's "labor process games" [17], willingly contributing massive amounts of data and creating surplus value for digital capital. In essence, within the "virtual production lines" painstakingly constructed by digital capital, app-based gig workers are merely mobile data points controlled by the platform system, busy filling the "AI last mile."
At present, since the advent of large model production, general agents are showing their prowess in an increasing number of knowledge-based jobs. In contrast, humanoid robots built on "embodied intelligence" are still only being showcased in laboratories, and the large-scale commercial use of autonomous driving and drone delivery still requires time. Furthermore, although app-based gig workers mostly engage in low-skilled manual labor, the agility and spontaneous decision-making they exhibit when completing tasks remain beyond the reach of even today's most sophisticated digital machines. Of course, in the long run, once "Moravec's Paradox" [18] in the development of artificial intelligence is overcome, "low-wage occupations in the era of the digital revolution may last for another decade or two, but soon their era will also come to an end."
In view of this, a topic more worth discussing in the short term is whether the app-based gig workers at the bottom of society can more decently "get a piece of the pie" in the feast of digital capital amidst the new wave of the digital revolution set off by large model production. The practice of resistance in capitalist digital labor shows that this fundamentally depends on whether the actual environment of gig labor can be continuously improved. From the perspective of platform empowerment, once general agents are embedded into platform algorithmic systems to become digital assistants for gig workers, laborers can be partially liberated from the suffocating "digital cage," experiencing more authentically the autonomy and freedom that digital labor can bring. Importantly, the sense of self-efficacy among gig workers exhausted by digital toil is enhanced by AI augmentation, which is not only conducive to improving labor productivity and promoting the production of surplus value, but also plays a role, to a certain extent, in alleviating and shifting the increasingly sharp contradictions between labor and capital. In response, shrewd digital capital need only extend the "olive branch" of "algorithms for good" [19] to rest more easily while pocketing the fruitful results extracted from gig labor.
4. The trend toward the "labor aristocracy" among employed digital workers intensifies A prominent feature of the labor system under digital capitalism is the dual employment system implemented by digital capital. On the one hand, for the vast number of general laborers, it relies on digital platforms to implement outsourcing and crowdsourcing to achieve "non-employment exploitation." On the other hand, to recruit technical geeks with immense innovative potential, digital capital competes to provide staggering high-salary packages through employment contracts, turning them into the "labor aristocracy" [20] of the digital age. Platform economy practice shows that digital capital strives to conceal its extreme extraction from non-employed workers while high-profilely advertising the excellent treatment and working environment of its own employees, using this to embellish its image and win the favor of capital markets and stakeholders.
Following the implementation of large model production, the plugin-based application of general agents will allow various professional digital tasks—such as web design and maintenance, coding, and data analysis, which were originally completed by formal employees—to be "done with one click" through natural and relaxed human-machine dialogue. For this reason, it is only logical that greedy digital capital has launched round after round of layoff plans over the past two years. At the same time, to win the competition in large model production, digital capital has used every trick in the book to compete globally for top talent in the development, optimization, and deployment of large models with even more enticing compensation packages. As a result, it is certain that whether in emerging digital tech enterprises or traditional enterprises undergoing digital transformation, the scale of formal employees will continue to shrink. However, senior digital laborers focused on the R&D, production, and application of general agents will further transition toward becoming a "labor aristocracy" amid the adulation of digital capital.
IV. The "Bostrom Question" and the Future Situation of the Movement of Value
Marx pointed out that in production based on capital, the production system mediated by exchange value achieves an "abstract domination" over people. Furthermore, as long as labor "is only posited as general labor through exchange," the capitalist movement of value will function as usual. After entering large model production, the substitution of capital for labor deepens, and the further infrastructuralization of digital means of production causes the movement of value in digital capitalism to face new situations and challenges. In the short term, the production agency of general agents and the AI augmentation of digital labor allow the capitalist digital production system to continue. But in the long run, once the development of general intelligence technology breaks through the "technological singularity," the shocking question raised by British scholar Nick Bostrom in his bestseller Superintelligence (i.e., the "Bostrom Question")—"Suppose that machine laborers could be produced very quickly and that they were cheaper and more capable than human laborers in all jobs, then what would happen?"—will likely no longer be mere somniloquy. Fundamentally, where the movement of value in digital capitalism goes depends on the ultimate evolution of the "human-like" and "machine-like" qualities of digital machine systems represented by general agents.
1. Digital imperialism and the continuation of the movement of value Under the "Turing Machine" paradigm, the general agent remains an "organ of the human brain, created by the human hand; the power of knowledge, objectified." However, as the Israeli scholar Yuval Noah Harari cautioned, "the belief that humans will always have a unique ability beyond the reach of non-conscious algorithms is just wishful thinking." Given this, if one day general agents without "awakened consciousness" comprehensively surpass human "labor power," and capital subsequently achieves the ultimate substitution of labor, then in the author's view, digital capitalism, hollowed out of its value substance, can only linger on through the illusory prosperity of a "digital empire."
First, regarding the institutional basis of the movement of value, under "Silicon Valley private ownership," the key infrastructure for large model production is controlled by a tiny handful of digital tech giants. It goes without saying that the privatization of human cognitive means of production drives the colonization of capitalist private ownership from physical space to digital network space and then to human cognitive space, further solidifying the foundation of the digital capitalist mode of production. Second, regarding the production basis of the movement of value, the factor of living labor in the digital capitalist production process will not completely disappear, but it is highly likely to shrink to the threshold of an "economic singularity." On the one hand, in the R&D, optimization, and deployment of general agents, the contribution of digital labor from top digital talent is huge, but after all, the input of living labor in the total sense is very small. On the other hand, under conditions where human labor is "preferred due to its aesthetic, ideological, moral, religious, or other non-pragmatic factors," handicrafts may be sought after by the market, but at best they are merely embellishments in an ocean of "machine-made commodities." Furthermore, regarding the market basis of the movement of value, future social production seems achievable through a few elites plus intelligent machines; large-scale technological unemployment for ordinary laborers will be inevitable. "People continue to depend on wages, yet wages become increasingly difficult to obtain." Consequently, faced with the sharpening "market paradox" of the movement of value in the era of large model production, it is not surprising that digital capital itself promotes "Universal Basic Income" (UBI) plans to create and maintain effective market demand.
2. AI capitalism and the metamorphosis of the movement of value It must be emphasized that the previous refutation of the "new worker" theory was based on the premise that the iteration of general agents does not involve the "awakening of consciousness." Theoretically, if the evolution of artificial general intelligence technology leads to "intelligence escape," where digital machine systems acquire "self-consciousness, reflexivity, and creativity" and dash toward becoming "Super-Turing Machines," then one might as well restate what the "new worker" theory describes: an "AI capitalist" society where general agents serve as the "labor subjects" will make its debut. Thus, under the AI capitalist mode of production, general agents become the new "wage laborers," digital capital further metamorphoses into "AI capital," and the "movement of value," which can only exist through indirect "forced labor," will function as before. By then, "the wage relationship is completely abolished; large corporate groups spreading across the world continue to own and control the means of production, but they no longer employ humans; the main task of humans is merely to provide data for the machines through daily activities."
Consequently, from the perspective of "value" formation, production based on "AI capital" finds new, devourable "living labor" and regains vitality; from the perspective of "value" realization, the "reproduction of labor power" by general agents continuously creates effective market demand, ensuring that the "overproduction" of AI capitalism does not easily spiral out of control; and from the perspective of the existence of "value," "AI capital" firmly controls the infrastructuralized key means of production, causing the institutional contradiction where "the production of use-value is limited by exchange-value" to recur. Therefore, it can be said that from the perspective of "wealth" production, this is a new world where "silicon-based producers" toil; from the perspective of the movement of "value," this is an old world still filled with "alienated labor" and "enslavement and exploitation." But in the final analysis, this is a "de-humanized" or even "anti-human" digital dystopia.
3. Digital socialism and the disintegration of the movement of value In Marx's view, the transformation of all products and activities into exchange value is a product of socio-historical development. Meanwhile, "the contradictions and antagonisms do not arise from the machinery itself, but from its capitalist application." When human society enters the stage of "the free and well-rounded development of individuals," and "on the basis of communal production, labor would be posited as general labor before exchange," the movement of value based on capital will naturally wither away. Furthermore, "in a communist society, the scope for the use of machinery would be entirely different from that in a bourgeois society." This will be prominently reflected in general agents truly becoming human digital assistants, liberating people from routine labor and allowing them to naturally and consciously engage in creative activities consistent with the "species-essence of man" [21].
Transcending the logic of capital and building digital socialism is the only way to ensure that large model production benefits all of humanity. To this end, we must abandon techno-utopian fantasies such as "post-capitalism," "fully automated luxury capitalism," and Western Leftist accelerationism; fundamentally transform the digital capitalist platform system; and establish and improve the basic economic system of digital socialism. It goes without saying that the most fundamental and urgent task is to promote the construction of "digital commons" and facilitate the opening of data corpora, the sharing of computing power infrastructure, and the open-sourcing of large model algorithms. This will lay a solid foundation for the mode of production to use large models as the "means of production for combined, social labor."
Supported by General Artificial Intelligence (GAI) technology, digital capitalism has entered the era of large models. The routine application of digital machine systems, represented by general-purpose intelligent agents, vigorously drives the generation and development of new quality productive forces. On one hand, this creates a digital wealth of prosperity for capitalism with high efficiency; on the other hand, it subjects the capitalist motion of value to major shocks. Faced with the revolutionary and subversive changes occurring under large model production across all aspects—including the subject of labor, the object of labor, the content of labor, the form of labor, and relations of production—it is a matter of great urgency to deeply advance the Sinicized, contemporary interpretation of Marx’s labor theory of value. Fundamentally speaking, only by upholding the fundamentals can we avoid losing our way, and only through innovation can we grasp and lead the times. To deepen our regularized [22] understanding of the laws governing the motion of value in digital capitalism, there is a pressing need to creatively apply the basic principles of Marx’s labor theory of value. This requires us to penetrate the complex productive appearances of labor-value creation practices in the era of large models, clarify the essence of digital machines, grasp the morphological changes of digital labor, analyze the intensification of contradictions and development trends in the motion of value, and seek a new form of human civilization [23] characterized by AI for the public good and the liberation of labor. In this regard, while adhering to the core idea that "living labor is the sole source of value," we must continue to resolutely strike back against various clichés which baselessly claim that Marx's labor theory of value is obsolete or invalid. At the same time, we must maintain theoretical vigilance and provide powerful refutations against various "new theories" that vulgarize, subjectivize, or de-humanize the creation of labor-value.
(About the Author: Huang Zaisheng is a professor in the Department of Marxist Theory at the Political College of the National Defence University) Online Editor: Tongxin Source: Marxism Studies (《马克思主义研究》), Issue 12, 2023.