Huang Zaisheng: Generative Artificial Intelligence and the New Domain of Value Movement in Digital Capitalism
In recent years, the high degree of innovation, strong permeability, and extensive coverage of a new generation of Artificial Intelligence (AI) technologies have exerted a broad and profound influence on the production, realization, and distribution of value in digital capitalism. In particular, the sudden emergence and dizzying iterative upgrades of generative AI—represented by ChatGPT—have greatly stimulated the value potential of data as a factor of production and promoted an overall leap in digital productive forces. This marks the entry of human knowledge production into the era of Large Models. The history of capitalist development shows that every major technological innovation and its derivative effects trigger adjustments in the relations of production and transformations in the production regime. At present, the production, exchange, circulation, and distribution of digital capitalism are undergoing accelerated intelligentization; "AI capitalism" has begun to take [initial] shape.
Within the horizon of Marxist political economy, the fundamental path through which capital, labor, technology, and institutions intermingle under the capitalist mode of production is one where capital exerts every effort to utilize technological progress to discipline labor in order to extract surplus value. Yet, simultaneously, the "subjectivization" of dead labor bolstered by technology inevitably feeds back to consume value. Facing the phenomenal craze triggered by ChatGPT "breaking the circle,"[1] a pressing question that demands attention and in-depth exploration is: in what "innovative" ways will the technological debut of generative AI and the comprehensive upgrade of intelligent machines enact the movement of value in digital capitalism?
To deepen our understanding of the laws governing the development of digital capitalism, it is necessary to carry out a preliminary exploration of the following cutting-edge issues. First, from the perspective of digital capitalist commodity production: generative AI is unlocking a new cognitive revolution; what new changes will this bring to the production and consumption of digital commodities? Second, from the perspective of the digital capitalist labor process: what new characteristics does the creation of value by digital labor manifest under human-machine collaboration? Third, from the perspective of digital capitalist capital accumulation: how will the competition among digital oligarchs over large models lead to new trends in digital capital accumulation? Furthermore, as AI technology accelerates, what new changes will the production order of digital capitalism face?
I. The Cognitive Revolution and a New Stage in the Production and Consumption of Digital Commodities
"The Taylorism of the 20th century touched off a 'productivity revolution' that rapidly increased labor productivity. In digital capitalist society, how to touch off a knowledge-based version of the productivity revolution—that is, a 'knowledge productivity revolution'—will become the key to development." Currently, the headlong rush of generative AI technology is initiating an industrial revolution in the field of human social knowledge production. The automation of capitalist information and knowledge production is gradually becoming a reality, thereby allowing the production of digital commodities to break through capacity constraints and accelerate toward a new stage of hyper-personalized customization.
(1) "Synthetic Data" Breaks the Supply Constraints of the Data Factor
In the era of the digital economy, data has become a key factor of production. In practice, after massive amounts of heterogeneous data undergo factor-processing—such as manual cleaning—they become the "fuel" for intelligent machines, the "raw material" for digital commodities, and the "assets" of digital platforms. However, the commercial practice of analytical AI in recent years indicates that the production and consumption of digital commodities are facing increasingly tight supply constraints regarding the data factor. The reasons for this include: the quality of existing data resources (text, images, audio, etc.) is uneven, making it difficult to cover long-tail and edge cases; data diversity is insufficient; the collection and labeling of AI training data is time-consuming, labor-intensive, and high-cost; and the uncertainty of data factor supply is rising due to factors such as government data regulation, user privacy protection, and digital geopolitics. Additionally, the deepening segmentation of the data commodity market has made the "shortage" of proprietary data in vertical domains more prominent.
Presently, AI large model technology is forging a new momentum for the valorization of data. Generative AI's extreme pursuit of and stunning performance in "human-like intelligence" cannot for a moment dispense with the "data feeding" of meticulously organized, high-quality corpora. According to reports, the AI language model GPT-3 was trained using 45 terabytes (TB) of data, roughly equivalent to 450,000 years' worth of text from the People's Daily. The training data for the conversational language model LaMDA came from 1.56 trillion words scraped from the internet. The Contrastive Language-Image Pre-training (CLIP) model collected over 400 million pairs of "text-image" training data from the internet. For digital capital, the "data hunger" is more violent than ever before. In response, digital tech giants are no longer satisfied with "data enclosure" and "data colonialism"; instead, they are taking an alternative path by actively developing synthetic data technology to efficiently achieve large-scale production of the data factor through data augmentation, simulated generation, and data collection in simulated environments.
Synthetic data can mathematically or statistically reflect the properties of real-world data, and can therefore serve as a substitute for real-world data to train, test, and validate AI large models. Furthermore, the cost of synthetic data is far lower than that of real data. Some organizations predict that by 2024, 60% of the data used to train large models will be synthetic, and by 2030, the vast majority of data used by AI large models will be synthesized by AI. Consequently, as synthetic data created through AI content generation technology becomes prevalent, it not only automatically generates standardized, pre-processed data—which is secure and labeled—to satisfy increasingly "hungry" large models, but the automation of data factor supply also allows digital capital to further escape the "fetters" of digital labor. This enables capital to drive intelligent machines more willfully at full speed to achieve the ultimate harvest of a new round of "data dividends" in the intelligent economy.
(2) Model Pre-training Prompts the "Generalization" of Intelligent Machine Productivity
Under the capitalist mode of production, "the means of labor pass through different metamorphoses, whose culmination is the machine, or rather, an automatic system of machinery."[2] Undeniably, having entered the stage of digital capitalism, the iterative upgrades of intelligent machines are demonstrating the "autonomy" of the "system of machinery" to the fullest extent with breathtaking technological acceleration. However, what has been "regrettable" for digital capital is that in the application scenarios of analytical AI, intelligent machines—appearing in concrete forms such as search engines, facial recognition, intelligent voice, and algorithmic recommendation—all focus on a single mode and specialize in pattern recognition and behavioral prediction for specific domains. To borrow a term from New Institutional Economics, these intelligent machines possess strong "asset specificity": "they might know what they’ve been taught, but they don’t know what they haven’t been taught." This "model island" phenomenon has greatly restricted the digital productive potential of intelligent machines.
By contrast, under the engineering combination of "Big Data + Large Computing Power + Strong Algorithms," AI large models can achieve the "miracle" of knowing "what they haven’t been taught." From the perspective of their manufacturing mechanism, digital capital adopts advanced neural network architectures to "pre-train" foundational models by "devouring" massive unlabeled datasets. Then, based on Reinforcement Learning from Human Feedback (RLHF), the models are fine-tuned. With only small-sample or even zero-sample learning, they can successfully complete new tasks and satisfy specific uses. That is, AI large models possess significant downstream task adaptation capabilities, promoting the "generalization" of intelligent machine productivity and achieving "industrial-style deployment across different industries, vertical domains, and functional scenarios." In other words, with the support of generative AI technology, the standardized development and ubiquitous application of intelligent machines can form a staggering "intelligent emergence" effect. This provides a virtual production line—installed on clouds and networks—for the automated production of capitalist digital commodities. Once the production of intelligent machines by intelligent machines becomes a reality, "AI capitalism" will possess its own characteristic means of production, thereby establishing a technical basis suited to itself.
(3) AI Content Generation Technology Prompts the Automation of Digital Content Production
In the stage of digital capitalism, digital content production has become the production of use-value most characteristic of the era in capitalist society. From the era of Web 1.0 Professionals Generated Content (PGC) to the Web 2.0 era of User Generated Content (UGC), and now transitioning to the current AI Generated Content (AIGC), the development and application of digital technology continue to drive transformations in digital content production modes.
As shown in Table 1, from the perspective of content production methods—whether it is text generation, script drawing, webpage translation, site creation, or code writing—AIGC both overcomes the "slow delivery" deficiency of PGC and eliminates the "mixed bag" [uneven quality] drawbacks of UGC, becoming a "factory" and "assembly line" for the automated production of digital content. This production mode can ultimately satisfy massive personalized demands in a high-efficiency manner with low marginal costs. Precisely for this reason, by virtue of its rapid response capabilities, vivid knowledge output, and rich application scenarios, AI content generation is increasingly becoming a new engine for human material and mental [spiritual] production.
Further, AI content generation technology empowers the development of various industries, spawning new products, business formats, and models in the digital economy, which further consolidates the value system of digital capitalism. First, in commercial applications such as text creation, human-machine interaction, audio-visuals, education, and finance, generative AI products can reach "human-like" or even "supra-human" levels. Their "utility of things" is astonishing, significantly enhancing user stickiness and consumption willingness. Second, "Model as a Service" (MaaS) has realized the transformation of AI applications from a "handicraft workshop" to a "factory mode." Under the MaaS business model, digital capital charges fees based on model invocation volume (i.e., according to the volume of data requests and actual computation), achieving pricing by "data/computation," which greatly reduces the market transaction costs of digital content commodities.
Moreover, after the emergence of generative AI technology, digital capital can not only use machine intelligence more easily to objectify and appropriate the "general intellect" of society, but it has also "infrastructuralized" intelligent machines to an unprecedented degree, keeping them firmly in its own hands. The result is that digital capital monopolizes the "means of cognition," further solidifying the ownership basis for the production of digital commodities.
(4) "Human-Machine Dialogue" Promotes the Upgrade of the Digital Consumption Experience
Since the birth of the computer, the "world of atoms" in human material production and life has been continuously "bit-ified." Increasingly rich digital products and services have profoundly changed the consumption landscape of the masses in contemporary capitalism. Especially after entering the Web 2.0 era, easy-to-operate and user-friendly Graphical User Interfaces (GUI) have largely eliminated the digital divide caused by "command line" operations; using machines and "going online" is no longer a geek trend, and ordinary people can easily enjoy convenient digital services. But fundamentally speaking, analytical AI displays information statically; its digital content presents an algorithmic "coldness" and technological alienation in human-machine interaction that cannot be dispelled. Furthermore, algorithmic manipulations such as "big data price discrimination"[3] frequently infringe upon user rights, further exacerbating the sense of user distrust in human-machine interaction.
Currently, the "breaking of the circle" by ChatGPT has accelerated the implementation of generative AI technology, driving an upgrade in the experience of digital commodity consumption. Based on the minimalist interaction of natural language, it allows ordinary users to instantly immerse themselves in a more natural, direct, and smooth conversational digital scenario, bringing an entirely fresh digital consumption experience. This is mainly reflected in the following aspects: regarding real-time information retrieval, generative search engines present information in the form of passage reading, replacing the simple summary list of links in traditional search engines. This puts information at the user's fingertips more quickly and comprehensively, greatly catering to the ordinary user's psychology of expecting a "second-response" answer. Regarding the supply of digital services, the embedded and plug-in integration of generative AI transforms the interaction methods of massive applications. Its functions cover fields such as online food ordering, corporate office work, process optimization, image and text processing, and product marketing and recommendation, realizing an enhancement of mobile software functions. Consequently, through precise reach that "understands what you think, knows what you need, and solves what worries you," and with the full participation of intelligent customer service, intelligent marketing, and digital human live-streaming, it brings a multi-sensory, interactive, and immersive experience that blends the virtual and the real to online consumption.
II. Human-Machine Collaboration and New Characteristics of Digital Labor Value Creation
With the dawn of the digital economy era, analytical AI [4] initiated capitalism’s “Second Machine Revolution” and gave rise to various forms of digital labor. Currently, technical applications of generative AI are mushrooming. "With the support of large models, AI is transforming from a 'tool' into a 'partner' for humanity." Human-machine collaboration will lead to the continuous deconstruction of digital labor tasks; meanwhile, both the methods and the content of digital work will undergo fundamental changes. Overall, the creation of digital labor value is exhibiting the new characteristic of "intelligence enhancement."
(1) Human-machine collaboration promotes the "value dimension-climbing" of unpaid labor The development of the platform economy demonstrates that active users are both consumers of digital services and producers of data and content for digital platforms. Regarding the human-machine relationship, the supportive role of analytical AI is primarily reflected in increasing the exposure and attention of user-generated content (UGC) through algorithmic recommendations, while its contribution to the actual production of user digital content has been lackluster. In other words, AI technological empowerment has actually been the "empowerment of the machine" [5], enabling ordinary users to achieve an orderly presentation of "genuine emotional flow" and "self-expression" within digital cyberspace. In the process of stimulating likes, comments, shares, and tips, this helps digital capital harvest massive amounts of data and user attention. In contrast, a series of easy-to-use generative AI applications like ChatGPT have greatly enriched the "toolbox" for user content generation. To be sure, "AI models will become the ubiquitous and capable assistants for every worker, putting new types of hyper-personalized intelligence into people’s hands to increase productivity." This is manifested in two ways under the human-machine collaboration model: on the one hand, instrumental tasks such as production, optimization, and personalization of UGC can be automatically completed by intelligent machines driven by generative AI technology; on the other hand, ordinary users are able to transcend the limitations of "technique" and "efficiency" to specialize in creative conception, content imagination, and self-expression. In this way, generative AI both "empowers the machine" and "empowers intelligence" [6]. Low-threshold AI creation makes "artistic existence" the "new normal" for ordinary people. Consequently, in terms of value creation, the content ecosystem of digital cyberspace is improved, the digital content economy develops healthily, and the value contribution of users’ unpaid labor becomes increasingly prominent.
(2) Human-machine collaboration triggers "market polarization" in crowdsourced labor In 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." Depending on the level of professional skill required to complete digital tasks, crowdsourced labor is further subdivided into freelance work, represented by Upwork, and micro-labor, represented by Amazon Mechanical Turk. In the online crowdsourcing economy, the "machine-empowering" effect of analytical AI is mainly reflected in breaking the spatio-temporal constraints of the labor market through algorithmic dispatching and algorithmic rewards/punishments, thereby either opening new channels for professionals to monetize "cognitive surplus" or providing new employment opportunities for marginalized and vulnerable groups.
With the accelerated development of embedded and plug-in applications of generative AI, the online crowdsourced labor market will undergo profound transformations, and the employment situation and value contribution of digital laborers will undergo a clear "market polarization." This is primarily because crowdsourced labor is essentially cognitive labor, and its procedural and repetitive tasks will sooner or later be automated. Among these, crowdsourced micro-labor mostly consists of low-threshold digital tasks that digital capital will certainly seize the opportunity to hand over to the "miracle tool for achieving ends"—intelligent machines. Existing research shows that compared to manual annotation, ChatGPT's automatic annotation of tweets (including relevance, stance, topic, and frame detection) is not only more accurate but also costs less than $0.003 per annotation—about 20 times cheaper than Amazon Mechanical Turk. For freelancers with "unique skills," although tasks with high repetition, regularity, and predictability in writing, translation, mapping, and code checking will shrink or even disappear, the labor returns for completing "unique" cognitive works will become more substantial. In practice, AI large models have derived a rich matrix of capabilities, helping digital creators break through productivity limits and thus dedicate more time and energy to more imaginative and challenging digital labor. According to Accenture’s forecast, by 2030, 75% of knowledge workers worldwide will interact daily with applications, services, or agents supported by foundational models. Furthermore, in human-machine dialogues, the massive knowledge content generated by intelligent machines under prompting and induction often includes "mind-blowing" "insights" or "creative inspirations." As a result, in the production of value under digital capitalism, the upgrading of intelligent machines supported by large models has, on one hand, led to the loss of job opportunities for ordinary laborers engaged in "ghost work," and on the other hand, increased wealth-creation opportunities for "digital nomads" who excel at creativity.
(3) Human-machine collaboration helps improve the "environment" of on-demand labor In the digital economy era, the gig economy has transcended spatio-temporal constraints and developed rapidly, giving rise to mobile app-based on-demand labor. In essence, within the virtual production assembly lines that digital capital has "painstakingly created," [7] on-demand gig workers are merely mobile data points controlled by platform systems, busy filling the "last mile of AI." From the perspective of human-machine relations, the technical support of analytical AI on one hand helps ordinary laborers achieve low-threshold rapid employment, promoting digital inclusivity and employment equity to a certain extent. On the other hand, platform enterprises aggressively implement "the strictest algorithms" and deliberately shed "employer responsibilities," causing many on-demand gig workers to fall into a state of physical and mental exhaustion and employment instability. Consequently, the scene once mentioned by Marx reappears in the digital economy era: namely, that "machinery, gifted with the wonderful power of shortening and fructifying human labour, we behold starving and overworking it."
In recent years, promoting "algorithms for good" and protecting the labor rights of on-demand gig workers have increasingly become a social consensus. Under pressure from government regulation and society, digital capital is also inclined to refine and improve algorithmic management through platform systems to appease the dissatisfaction of digital laborers and consolidate the platform production regime. The integration of generative AI into platforms can push the gig economy from human-machine opposition toward human-machine collaboration, thereby improving the experience of on-demand labor and easing contradictions between labor and capital. This is because the intelligent upgrading of platform digital customer service—transforming from "cold" and procedural responses to "friendly and relaxed" chat dialogues—even allows workers to customize "bosom partners" or "AI assistants." This upgrade can, on one hand, create a more pleasant and humane digital labor environment for workers; on the other hand, it can adapt to different scenarios, providing workers with more personalized service prompts, process planning, and safety reminders in real-time, making the task completion process "effortless." Furthermore, as native applications of generative AI become increasingly rich, laborers can also freely create "digital twins" to maintain communication with multiple clients simultaneously, better catering to consumer preferences.
(4) Human-machine collaboration drives "job transformation" for employed digital labor Under the platform production regime, digital capital pushes flexible employment practices to the limit. On one hand, through the construction of multilateral market relations, it achieves "non-employment exploitation" of digital labor in the forms of online crowdsourcing and outsourcing. On the other hand, digital capital owners spare no expense to recruit global digital technology talents to engage in platform system infrastructure design, algorithm development, and digital product R&D and operations, making them enterprise architects, algorithm engineers, data analysts, and O&M engineers. These digital technology talents hold multiple identities: they are highly educated mental laborers, digital industrial workers, and members of middle-to-high-income groups. Broadly speaking, in the application scenarios of analytical AI, the production contribution of these employed digital laborers is manifested in the data valorization chain of "analyzing data, discovering patterns, forming insights, and establishing predictions."
With the accelerated arrival of the AI large model era, the automation of digital commodity production tasks is moving to a higher level. Driven by generative AI technology, the focus of technical geeks gathered in leading platform enterprises will shift, and new job positions will emerge. Specifically, in platform infrastructure construction, seeking new breakthroughs in areas such as new chipset architectures, hardware innovation, and efficient algorithms has become a key task for core employees of digital tech giants. In AI large model R&D, the interlinked stages of model construction, pre-training, fine-tuning alignment, and inference deployment have made professions such as algorithm engineers and prompt engineers popular digital occupations. In vertical applications, work such as training, fine-tuning, pruning, and distillation of industry-specific models has given birth to new digital labor professions such as AI product managers, AI quality controllers, and AI ethics consultants. In short, specializing in the R&D and production of artificial general intelligence (AGI) machines is increasingly becoming the daily work routine for digital laborers.
III. Large Model Competition and New Trends in Digital Capital Accumulation
Marx pointed out: "A certain stage of capital accumulation is a condition for the specifically capitalist mode of production, and the specifically capitalist mode of production in its turn causes an accelerated accumulation of capital." Entering the stage of digital capitalism, analytical AI has driven the platform revolution, giving rise to and consolidating "data enclosures," algorithmic despotism, and platform monopolies, achieving the accelerated accumulation of digital capital in a very short time. Currently, with the upgrading, iteration, and accelerated implementation of generative AI, capitalist production has entered the era of AI large models. The AI large model industry is technology-, capital-, and talent-intensive. As shown in Table 2, competition in large models is consolidating the digital oligopoly structure, causing the accumulation of digital capital to exhibit a new trend of "intelligent computing monopoly" [8].
(1) The struggle for computing power becomes the new focus of digital capital accumulation In the digital economy era, the "trinity" of data, computing power, and algorithms has become the necessary resources for digital tech giants to seek and consolidate platform monopolies. With the application of AI content generation technology in the production of digital commodities, digital platforms using advanced algorithms as competitive weapons are accelerating their advancement toward high-performance computing platforms. In the new round of resource competition by digital capital, influenced by factors such as the diffusion of the platform production regime, the surge in open-source databases, and the application of synthetic data, the importance of data resources has begun to decline. In contrast, creating foundational models is a complex, costly, and computationally intensive task; the amount of computation required to train top-tier AI models is growing exponentially.
Taking the AI language model GPT-3 as an example, this model has 175 billion parameters and utilizes a high-performance computing cluster consisting of 10,000 V100 GPUs and 285,000 CPUs. A single training run takes 14.8 days, with a total computing power consumption of approximately 3,640 PF-days (i.e., if one quadrillion calculations are performed per second, it would take 3,640 days). This means that in the digital productivity revolution driven by generative AI technology, the importance of intelligent computing power used for training, tuning, and deploying large models has become acutely prominent. It can be said that the competition in AI large models is, in reality, a competition for computing power; whoever possesses abundant advanced computing power will have a high probability of total victory in the large model competition.
(2) Competition among digital oligopolies drives up the monopoly over digital means of production The movement of value in capitalism "presupposes the divorce of the laborers from the ownership of the conditions for the realization of their labor." "As soon as capitalist production is once on its own legs, it not only maintains this separation, but reproduces it on a constantly extending scale." From a practical path, capital strives to promote the machine revolution, using "objectified power of knowledge" to achieve the "deskilling" of labor and the predatory appropriation of social "general intellect." Entering the stage of digital capitalism, the establishment and consolidation of the platform production regime have pushed the monopoly of digital means of production to new heights.
At present, although it seems like "a hundred boats are racing" in the field of large models, the engineering combination of "big data + big computing power + strong algorithms" means that the real players are few and far between. "Except for the world's top enterprises, almost all organizations are unable to complete this task on their own; it exceeds the capabilities and methods they possess." Therefore, although "AI startups" are everywhere, they can only become "barnacles on the hulls" of big tech companies.
To seize business opportunities in the AI "arms race," digital oligarchs are racing to restructure platform ecosystems. On one hand, these oligarchs are applying generative AI to every self-operated product, service, and business process. In March 2023, Microsoft integrated generative AI technology into its product matrix and launched Microsoft 365 Copilot; it can not only directly generate Word documents, PPTs, and Excel spreadsheets based on brief user commands but also assist users in completing tasks such as organizing meeting summaries and processing emails. Google followed closely behind, integrating an upgraded version of Bard into its "family bucket" of office software and creating an office assistant named "Duet AI." On the other hand, by building generative AI "super apps," digital oligarchs are embedding large AI models as core infrastructure into numerous industries, including health, medical care, education, logistics, credit, and culture and entertainment. Consequently, digital capital has not only further strengthened its dominance and control over the platform ecosystem but has also pushed the monopoly of the means of production—which sustains the capitalist mode of production—to the extreme through the oligarchic repositioning of cognitive tools of production. The result is that, from the perspective of stages of economic development, digital capitalism is accelerating its evolution toward "AI capitalism," and the dependence of entire social production and life on platforms will increase day by day.
(3) Large Models Create the "Surplus Proletariat" Under the capitalist mode of production, "as soon as the means of labor appears as a machine, it immediately becomes a competitor to the worker himself." During the period of industrial capitalism, the capitalist machine revolution progressed through steam power, electrification, and informatization; the devaluation and replacement of labor by machines (systems) remained the core issue of capital’s domination over labor. Overall, however, technological progress continuously birthed new jobs and new forms of labor, ensuring that the development of the capitalist relative surplus population never spiraled "out of control."
Following the emergence of AI technology, the effects of "labor enhancement" and "labor replacement" have coexisted. Currently, as generative AI technology accelerates its commercial application and industrial development, the technological unemployment caused by "machines replacing humans" seems to be taking on an overwhelming momentum. In the short term, although the technological empowerment and human-machine collaboration of generative AI have promoted the high-level evolution of digital labor as mentioned earlier, it must also be noted that a large number of ordinary laborers, unable to adapt to the accelerated changes in digital technology, are falling into the ranks of the "surplus proletariat." In the long run—even setting aside discussions of "super-intelligence," "social exclusion," and the "useless class" [9]—under the onslaught of rapid advances in generative AI, tasks or work exclusive to human knowledge creation will be continuously eroded by increasingly powerful intelligent machines. As some scholars have noted, while generative AI is a "powerful tool for labor, it may also become a 'monster' that 'devours' human digital labor." The result is that intelligent production is sharpening the problem of "technological unemployment," placing ordinary laborers such as the "digital poor" in a more challenging and hopeless situation, which will deliver a major shock to the digital capitalist production order.
IV. AI Technological Acceleration and New Variables in the Capitalist Production Order
"Constant revolutionizing of production, uninterrupted disturbance of all social conditions, everlasting uncertainty and agitation distinguish the bourgeois epoch from all earlier ones." From the perspective of the capitalist application of science and technology, "the impact of technological acceleration on social reality is undoubtedly enormous." Entering the era of the digital economy, the progress of digital technology driven by capital exhibits a "flywheel effect," making the capitalist production order more full of variables, turbulence, and even crisis. Currently, the emergence and iteration of generative AI have made capitalist technological acceleration even more unstoppable. Digital capital pursues "Artificial General Intelligence" (AGI), yet this simultaneously brings new variables to the capitalist production order.
(1) AI Technological Acceleration Constrains the "Spatio-Temporal Fix" of Digital Capital Marx pointed out that capital, on one hand, "calls to life all the powers of science and of nature... in order to make the creation of wealth independent... of the labor time employed on it," while on the other hand, it "wants to use labor time as the measuring rod for the giant social forces thereby created." Since the first Industrial Revolution, capital has successfully absorbed a large amount of "living labor" through successive "spatio-temporal fixes" [10] to hedge against the impact on capitalist rule caused by the antithetical development of the "expansion of the wealth system" and the "atrophy of the value system." Entering the era of the digital economy, digital capital has constructed a platform production system, greatly expanding its capacity for a "spatio-temporal fix." In sum, digital capital disciplines digital labor to realize the absorption of "living labor" on a larger scale; simultaneously, digital capital promotes "digital existence" to more thoroughly capture the social "general intellect." [11]
At present, the accelerated application of generative AI technology has released immense digital productive forces, creating an abundant flow of digital wealth with stunning efficiency. Yet, at the same time, it realizes a more thorough replacement of living labor with "dead labor" at an unimaginable speed. Thus, paradoxically, the more developed the digital capitalist productive forces become, the more the nightmare of the "approaching singularity" [12] in the movement of capital value becomes inescapable. Furthermore, under the action of the coercive laws of competition, as the application of large AI models becomes low-code and popularized, the organic composition of capital across the global knowledge industry is rapidly converging. Consequently, the "spatio-temporal fix" capacity of digital capital to build "digital empires" and to exploit and transfer surplus value is also being continuously dissipated. The result is that digital capital is mired in an AI-driven digital vortex; its blood-sucking capacity for "sucking living labor"—the very basis of its existence—is instead accelerating its own demise through its obsession with digital technology.
(2) AI Technological Acceleration Shocks the Realization of Value for Knowledge Commodities It is well known that under market economy conditions, knowledge commodities such as software, databases, and product designs possess significant zero-marginal-cost characteristics, which introduce more variables into their value realization. The creation of the capitalist intellectual property system effectively rejects various "anti-value" forces, thereby playing a key role in maintaining the capitalist production order. In the digital economy era, the self-valorization of digital capital is predicated on the deliberate dissolution of the personal intellectual property rights of user-generated content (UGC). The real-time, personalized, and scenario-based nature of digital commodity production allows digital capital to profit handsomely even without the protection of the copyright system. Overall, however, even with the support of analytical AI technology, digital capital has not broken through existing interest patterns to apply "de-copyrighting" to professionally generated content (PGC); thus, it has not yet caused a fundamental shock to the realization of value for capitalist knowledge commodities.
In contrast, the prevalence of generative AI is impacting the existing capitalist copyright legal system with lightning speed. This is because large AI models, by virtue of their increasingly powerful learning capabilities, have essentially netted and "devoured" all the intellectual achievements of humanity throughout history, producing an infinite amount of "non-authentic yet realistic" AI-generated content (AIGC) with extreme efficiency. For instance, while a human illustrator needs at least two to three days to complete a work, an intelligent drawing tool like Midjourney requires only ten seconds to create four works. Faced with a "sea of machine-generated text" that feels familiar yet is spurious, the current capitalist intellectual property system appears powerless and even at a loss regarding the tracing of origins, the determination of rights, and the identification of infringement. Generative AI can also create AI-cloned personas through style imitation, voice synthesis, and pose synthesis, accumulating users and harvesting traffic in a very short time with an overwhelming supply of creative digital content. Thus, under the impact of massive AI-driven mimicry and traffic diversion, the intellectual property value of original rights holders and real-world content producers is being greatly diluted.
(3) AI Technological Acceleration Aggravates Conflicts of Power Between Government and Enterprise Power is based on the control of information and its dissemination. Entering the digital economy era, digital capital has accumulated platform power through increasingly sophisticated algorithmic systems, building "digital empires." Practice shows that with the formation and consolidation of platform monopolies, the exercise of digital capital's power is no longer limited to the sphere of market exchange. Instead, through digital surveillance and information manipulation, it penetrates every aspect of human production and life, quietly playing the role of a "private government" and appearing as a "Leviathan" within platform society. "Platform companies, as subjects of power, have gradually moved to the forefront, dissolving and restructuring the state-led power structure, accelerating the dissipation of power from the state to the market." Therefore, when faced with the "usurpation of power" inadvertently achieved by digital oligarchs, bourgeois governments, once they wake up to it, become doubly vigilant and brew a counterattack. Fundamentally, in the game between government and enterprise in digital capitalism, digital capital’s acquisition of power stems from the infrastructurization of digital media. Since the emergence of the platform revolution, supported by the application of analytical AI, the generation of digital capital's power began with linking the supply and demand of multilateral markets and manipulating the precise distribution of digital content. Furthermore, the power and influence of digital capital are reflected in its ability to sway the online behavior of market subjects through platform "empowerment" and to influence user experience and psychological cognition through personalized pushes. Moreover, the active users submitting to "digital empires" often number in the hundreds of millions, willingly acting as "digital sharecroppers," which means the "governmentalization" of digital capital's private power should not be underestimated.
Practice shows that for the capitalist production order to maintain "smooth" operation, it is inseparable from the government’s use of public power to construct "social embedding" mechanisms [13] to mitigate labor-capital contradictions. Currently, with the accelerated implementation of generative AI, digital capital has greatly consolidated the power base of its "digital empire" by redefining user experience and reshaping the platform service ecosystem. More critically, digital capital can not only firmly control the diffusion of intellectual content through intelligent dissemination but can also directly acquire "new power" in the shaping of values and the production of ideas. That is, digital capital "sets its own format, standards, and value orientations for knowledge production through the design of models and the selection of corpora, becoming the controller of knowledge production." Thus, through the output of natural and fluent AI-generated content, digital capital subtly infuses its "private goods" of profit-seeking concepts, while the social masses, obsessed with the AI experience, are manipulated and persuaded without knowing it. As some scholars have pointed out: "Once the two major capabilities of AI in content production and content distribution form a closed loop, the dominance over information flow and the control over consumers will be immensely powerful." In this way, digital capital can provide information and knowledge through hyper-personalized customization, seizing the rights of interpretation and discourse, reshaping the capitalist knowledge order, and further eroding the ideological power of the bourgeois state, hollowing out the government's "toolbox" for soft social governance and threatening the capitalist political order. For this reason, it is only to be expected that the struggle for influence between government and enterprise in digital capitalism will intensify.
V. Conclusion With the iteration and everyday application of generative AI such as GPT-4, Gemini, and Claude, human production and life have entered the era of large AI models; the production process and relations of production of digital capitalism are undergoing revolutionary and subversive changes. Overall, the implementation and production maintenance of generative AI are initiating a new industrial revolution, as the mental production of digital capitalism enters a new stage of "hyper-personalized customization and one-click generation." At the same time, the arrival of the era of large AI models has accelerated the popularization of AI technology, and human-machine relations in digital capitalism are undergoing complex and frequent changes. However, the increasing "human-like intelligence" and autonomy of intelligent machines have not ended digital labor; on the contrary, human-machine collaboration is liberating people from repetitive and tedious labor tasks to focus on more creative work content. Furthermore, under "Silicon Valley private ownership," generative AI has, on one hand, efficiently created a boom in digital wealth, and on the other hand, it has intensified platform monopolies, relative surplus population, and conflicts of power between government and enterprise, making the inherent contradictions in the movement of value in digital capitalism increasingly difficult to reconcile. Only by transcending the logic of capital, thoroughly transforming the digital capitalist platform system, and exploring the construction of digital socialism can we truly realize the universal benefit of AI and the liberation of digital labor.
(Affiliation: Political Academy of the National Defense University) Web Editor: Tongxin Source: Theoretical Trends Abroad (国外理论动态), Issue 1, 2024