Accelerating Development of New Quality Productivity Empowered by AI Models

New quality productivity^① is a new engine for economic growth and a new driving force for high-quality economic development. In the digital intelligence era, general artificial intelligence (AGI) is an important driving force for the rapid development of new quality productivity. The accelerated iteration and daily application of AI models represented by ChatGPT (a chatbot program developed by OpenAI) are opening up a new industrial revolution^[1], promoting the leap of productivity from computing power to machine intelligence, and pushing the network-driven digital economy to advance towards a data-driven intelligent economy. Fundamentally, the intelligent productivity formed by the application of AI models is an important manifestation of new quality productivity. In this context, analyzing the internal mechanisms of AI models empowering the rapid development of new quality productivity from the perspective of Marxist political economy, clarifying the real obstacles to the rapid development of new quality productivity empowered by AI models, and pointing out the practical approaches for the rapid development of new quality productivity empowered by AI models are of significant theoretical and practical importance for firmly grasping the new technological revolution wave such as AI under the unprecedented changes in the world, deepening the understanding of the laws of productivity development in the digital intelligence era, and continuously strengthening and optimizing China’s intelligent economy.

1. Internal Mechanisms of AI Models Empowering Rapid Development of New Quality Productivity

Productivity is the most active and revolutionary factor, and it is the ultimate determining force driving the continuous evolution of human civilization. In the digital intelligence era, the new type of digitalized and intelligent infrastructure built by AI model technology is having a disruptive impact on human production and lifestyle. For China, new quality productivity is significantly supported by new technologies of digitization, networking, and intelligence. The implementation and large-scale application of AI model technology facilitate the deep integration of the digital and physical realms, thus providing unprecedented historical opportunities for the rapid development of new quality productivity.

(1) AI Models as General Purpose Technologies in the Digital Intelligence Era

Throughout human history, the emergence and diffusion of general purpose technologies (GPT) have continuously driven the transformation of human production methods and the high-level evolution of human civilization. From the perspective of technological-economic paradigms, farming technology nurtured the agricultural economy, steam engine technology and electrical technology supported the industrial economy, information technology gave birth to the information economy, and digital technology has created the digital economy. At present, AI model technology represented by ChatGPT possesses significant features of strong intelligence, strong generality, and strong interactivity, and is becoming a key foundational base for shaping human intelligent economy. As a report states: “Large models leverage the technological advantages of big data, big computing power, and multimodal capabilities to achieve a leap from perceiving the world, understanding the world, to creating the world, ushering human society into the intelligent societal stage of human-machine symbiosis.”^[2]

Currently, GPT-4 (the language model of ChatGPT), Gemini (a multimodal AI model launched by Google), Claude (a large language model released by Anthropic), Wenxin Yiyan, iFlytek, and other AI models are pre-trained on massive data based on powerful computing resources, and after prompt tuning, they can complete various downstream tasks to meet specific use cases. As the “pioneers” of AGI, the “disruptors” of content production methods, and the “collaborators” in human-machine interaction, AI models exhibit characteristics such as universality, progressiveness, and innovation-breeding potential as required by GPT, making them highly likely to become a new form of GPT^[3]1. This is reflected in the following two aspects.

On one hand, from the perspective of technological advancement, AI models, as an integration of generative AI technology and machine representation, mark a milestone in the development of AGI. Compared to decision-making AI, AI models efficiently mine vast amounts of multi-source heterogeneous data under the engineering combination of “big data + big computing power + strong algorithms,” giving rise to intelligent emergence and showcasing features and advantages such as autonomous learning, cross-modal understanding, reasoning, abstract thinking, and understanding of human society, thereby achieving a technological leap from “data flywheel” to “wisdom flywheel,” from simple perception to content creation, and from “single champion” to “all-rounder.” The capabilities of AI models are no longer limited to specific areas such as natural language and vision, but they possess the ability to perform general intelligent behaviors, widely expanding the applicability of AI^[3]2. On the other hand, from the perspective of technological application, with the enhanced availability of generative AI technology and the improvement of industrial informationization levels, AI models are triggering a new round of cognitive intelligence revolution, and with breakthroughs in data perception, intelligent cognition, and dynamic decision-making, they are becoming an important driving force for promoting technological leapfrogging, industrial optimization, and the rapid development of new quality productivity. Specifically, in the procurement sector, AI models focus on the product selection stage, conducting precise analysis of complex demands through dialogue-based business intelligence (BI) and making professional product recommendations, thereby maximizing procurement efficiency; in the manufacturing sector, AI models adapt to scenarios or form external plugin tools, pushing the system reconstruction and comprehensive upgrading of intelligent manufacturing through multi-format data comprehensive transformation analysis and PLC code generation; in the R&D sector, AI models target specific scenarios such as industrial design and drug development, expanding innovative boundaries, reducing innovation costs, and shortening R&D time through applications like computer-aided design (CAD) generation^②; in the marketing sector, AI models leverage vast customer service process data to create an “intelligent customer service brain,” using natural language Q&A interaction and powerful semantic generation capabilities to transcend traditional mechanical Q&A matching, opening up new paths for customer service “intelligent transformation.” Meanwhile, intelligent generation of marketing copy, virtual stores, digital human live streaming, and intelligent product selection can provide more efficient and personalized user experiences.

(2) AI Models Driving the Qualitative Change of Production Factors

In the context of political economy, productivity refers to the ability to utilize laborers, labor materials, and labor objects to produce products and provide services, which is specifically manifested in the process of laborers applying labor materials to labor objects^[4]. The history of human social and economic development shows that every leap in productivity is a practical product of qualitative changes in production factors driven by major technological advancements. In the agricultural era, land and labor were the primary production factors, and the emergence and improvement of hand tools drove the slow advancement of agricultural productivity; in the industrial era, in addition to land and labor, capital, technology, and management also became major production factors, and the application of machine systems and innovations in organizational models drove the rapid development of industrial productivity; in the information era, the production role of information factors became prominent, and breakthroughs in information technology and the development of emerging industries led to exponential growth in information productivity. In other words, the development of productivity ultimately boils down to the social nature of the labor being performed, the division of labor within society, and the development of intellectual labor, particularly in natural sciences^[5].

Entering the digital intelligence era, the factorization of data is continuously advancing, and the digitalization of industries and the industrialization of digital technologies are nurturing and promoting the rapid changes in digital productivity. The accelerated implementation and promotion of AI model technology are further releasing the value of data elements in a world defined by “data + computing power + algorithms,” promoting the upgrading and iteration of intelligent machines, and empowering laborers with AI, thereby driving an overall leap in intelligent productivity through the qualitative change of production factors.

1. AI Models Shaping New Types of Laborers

Laborers are the most active and decisive agents in productivity. In the digital intelligence era, a large number of repetitive, procedural physical and mental labor tasks are increasingly being replaced by ever-advancing intelligent machines. However, at the same time, various forms of digital laborers such as data scientists, chief data officers, data analysts, data compliance officers, digital engineers, and data brokers are emerging. With the reality of AI model production, intelligent machines begin to simulate the “spiritual production” of humans, redefining the division of labor in the production process between humans and machines, and human-machine collaboration is becoming the new norm for wealth creation, leading to the emergence of new types of laborers that create new quality productivity. Currently, generative AI applications such as ChatGPT, DaLL·E (an image generation model developed by OpenAI), Stable Diffusion (an image generation model developed by the CompVis research group at the University of Munich), and GitHub Copilot (an AI programming plugin developed by GitHub) have greatly enriched the “toolbox” for knowledge workers. It is certain that AI models will become ubiquitous and effective assistants for every worker, placing new personalized intelligence in people’s hands to enhance productivity^[6]. This is reflected in: with the help of the capability matrix derived from AI models, on one hand, repetitive, rule-based, and highly predictable tasks such as writing, translation, tabulation, graphic design, and code verification can be automatically completed by intelligent machines driven by AI model technology; on the other hand, new types of laborers break through the limitations of “skills” and “efficiency,” focusing on solving more imaginative and challenging digital labor issues. For example, for media practitioners, utilizing AI-generated content (AIGC) for part of the labor-intensive editing work allows them to concentrate more on news features, in-depth reports, and special reports that require deep thinking and creativity^[7]. Accenture predicts that by 2030, 75% of global knowledge workers will interact daily with applications, services, or agents supported by foundational models^[8]. As a result, laborers will no longer be trapped as tools in cubicles or on assembly lines, but will become proactive owners who flexibly respond to market demands and continuously innovate^[9]. More importantly, with the iterative evolution of general intelligent agents capable of autonomously planning and executing tasks, the concept of a one-person company with “individual + AI” is no longer a fantasy, and super individuals who are adept at “generating ideas” are already emerging^③.

2. AI Models Creating New Dynamics for Data Value

New quality productivity regards data as a key production factor. Essentially, the new generation of AI relies on a “dual drive” of data and knowledge; the more data available, the smarter the system becomes^[10]. From the working mechanism of intelligent machines, the breakthrough of generative AI technology is accelerating AI’s transition from being “model-centered” to “data-centered”; the AI model’s extreme pursuit of “human-like intelligence” is inseparable from high-quality data “feeding.” More importantly, as human-machine collaboration deepens, the autonomy of intelligent machines continues to increase, and the requirements for the timeliness, accuracy, and completeness of data are constantly rising, highlighting the value of high-quality data elements. Moreover, while synthetic data generated by AIGC can be applied on a large scale in fields such as autonomous driving, smart healthcare, and digital finance, the multimodal data that accurately records human production and life embodies the “general intelligence” referred to by Marx, directly impacting the “intelligence upgrade” of AI models and the upper limits of intelligent machines’ capabilities. Thus, compared to the commercial intelligence notion of “no data, no prediction,” in the increasingly expansive application scenarios of AI models, the principle of “no data, no generation; no data, no decision” more prominently illustrates the value release of data elements. In summary, AI models breathe new life into data, and the value of massive data will further emerge.

3. AI Models Giving Rise to New Production Tools

Marx pointed out: “The starting point of large industry is the revolution of labor materials.”^[11] Entering the digital intelligence era, intelligent machines are an important sign of the development level of new quality productivity. In recent years, AI has accelerated its transition from perceptual intelligence to cognitive intelligence, continuously expanding the capability boundaries of intelligent machines. However, generally speaking, in the application scenarios of analytical AI, intelligent machines that manifest as search engines, facial recognition, intelligent voice, and algorithm recommendation possess strong “asset specificity,” focusing on “listening, speaking, and seeing” in a single modality, specializing in pattern recognition and behavior prediction in specific fields. In contrast, driven by generative AI, AI models learn common knowledge from massive data, moving towards “thinking, creating, making decisions, and reasoning” with strong generalization, deep structure, and adaptive algorithms, and are being rapidly toolized under the guidance of industrial demand. Essentially, intelligent machines driven by AI model technology still present as objectified knowledge forces in the form of fixed capital. However, general intelligent agents represented by AutoGPT (an open-source application on GitHub) participate in the labor process in a more “human-like” manner, relying on their continuously evolving capabilities of perception, memory, understanding, analysis, and generation. Therefore, more accurately, intelligent machines driven by AI model technology are “production agents under human supervision”^[12]. It is certain that with the accelerated iteration of AI model technology and innovative scene applications, general intelligent agents are expected to accelerate the development of various industries, becoming new foundational production tools in the digital intelligence era.

(3) AI Models Driving the Improvement of Total Factor Productivity

New quality productivity is marked by the improvement of total factor productivity. Currently, AI model technology is empowering various industries, profoundly changing production functions and methods, and containing the enormous potential to “ignite a knowledge-based productivity revolution”^[13]. In other words, the powerful general intelligence possessed by AI models is showing enormous transformative power across industries, making AIGC demonstrate a “multiplicative power” effect in different application fields^[14]. As some scholars have pointed out, AIGC can gradually achieve a virtuous cycle of continuously optimizing resource allocation and forming innovative technologies driven by the three elements of “data + computing power + algorithms,” thereby promoting the improvement of total factor productivity in the process of enhancing productivity and technological progress^[15].

Firstly, from the perspective of content generation methods, AIGC automatically creates text, audio, images, videos, and 3D models, breaking through the upper limits of creative content and volume, with “super personalized customization and one-click generation” making content production efficient, easy, and rich in personality, increasingly becoming a new engine for human knowledge production. Secondly, from the perspective of human-machine collaboration, new quality productivity emphasizes high collaboration between humans and machines, pursuing “human-machine symbiosis” and “human-machine co-creation”^[16]. In this regard, the autonomy of general intelligent agents empowered by AI model technology is increasingly enhanced, enabling them to integrate knowledge from different fields and become effective digital assistants, freeing people from repetitive and tedious tasks to focus on more creative work. Research shows that software developers using Microsoft GitHub Copilot complete tasks 56% faster than those who do not use the tool^[17]. Gartner’s report predicts that by 2026, over 100 million people will work alongside “robot colleagues”^[18]. Thirdly, from the perspective of production operations, AI models can provide intelligent solutions in core scenarios such as production planning and scheduling, process parameter optimization, operational process monitoring, energy consumption governance, safety warnings, and product quality inspection, based on the interconnection of massive production factors, the value mining of operational data, and the sedimentation and reuse of industrial knowledge. Moreover, they can achieve leaps and improvements from “smart production lines” to “smart enterprises” and then to “smart industry chains” through comprehensive linking across the entire industrial chain, value chain, and all factors. Fourthly, from the perspective of R&D innovation, AI models, with their powerful content generation capabilities, further enhance digital twins, enrich the digital world, and promote the optimization and efficiency of simulations. Simultaneously, under the collaborative innovation of “machines + humans + networks,” general intelligent agents can greatly expand the knowledge production boundaries, helping humans explore unimaginable solutions in a short time using more variables. The Internet Data Center (IDC) predicts that by 2025, 35% of enterprises will master the methods to use generative AI to develop digital products and services, achieving revenue growth that is twice that of competitors^[14].

(4) AI Models Giving Birth to Future Industrial Patterns

Accelerating the development of new quality productivity and developing future industries are key. This requires targeting new directions, laying out new fields, and opening up new tracks, relying on original, cutting-edge, and disruptive technologies to nurture new industries and empower the transformation and upgrading of the entire industrial system. In this regard, as a cutting-edge technology of AGI, AI model technology is becoming an important driving force for transformation across various industries^[19]. Whether it is nurturing new industries, new models, and new business formats, or promoting the transformation and upgrading of traditional industries, AI models contain enormous potential for wisdom.

1. AI Models Giving Rise to New Business Formats in Intelligent Services

Under the support of foundational models, user-defined GPTs will give rise to a vast number of AI-native applications^④. Natural conversational user interfaces (CUI) significantly enhance the usability of AI. Given time, new models and business formats such as AIGC content creation, AI assistants, digital humans, marketing tools, and chat assistants will experience explosive growth. At the same time, AI models, as core foundational infrastructure, are embedded in numerous industries such as health, medical care, education, logistics, credit, and entertainment, forming super applications and injecting “intelligent elements” into the platform economy ecosystem. According to monitoring data from data.ai, in 2023, global downloads of generative AI applications increased ninefold, AI chatbots grew 72 times, and applications with embedded AI functionalities saw a 60% increase in downloads^[20].

2. AI Models Promoting the Intelligent Transformation and Upgrade of Traditional Digital Services

In March 2023, Microsoft integrated generative AI technology into its product matrix, launching the AI assistant—Microsoft 365 Copilot, which can automatically generate documents, create professional spreadsheets, optimize PPT layouts, and handle emails. In June of the same year, Microsoft launched Windows Copilot, which integrates sidebar tools of the operating system to help users complete various tasks such as answering questions, summarizing information, editing documents, and adjusting computer settings.

3. AI Models Opening New Growth Spaces for Intelligent Hardware through Edge Deployment

With the maturity and application of personal model technology, advanced AI-capable devices such as AI PCs, smartphones, smart speakers, smart cars, and humanoid robots will “fly into ordinary households,” greatly enhancing user experience and becoming standard equipment in people’s daily work and life.

4. AI Models Assisting the Intelligent Transformation of Traditional Industries

AI models targeting vertical fields are fusing with industrial internet, continuously enhancing industrial visual intelligence, industrial data intelligence, and industrial interaction intelligence, promoting the deep integration of digital technology with the real economy. Leading enterprises in the market are using private domain knowledge to build industry models, exclusive models, or scenario models, developing digital twins, industrial design, drug simulation, power grid modeling, video generation, animation rendering, and other businesses; small and medium-sized enterprises can rely on cloud intelligence platforms formed by “IaaS + PaaS + MaaS + SaaS” to enjoy model services that empower intelligence through data utilization.

2. Real Obstacles to Rapid Development of New Quality Productivity Empowered by AI Models

Currently, the new generation of AI development in China is in full swing, and the engine effect of AI models empowering new quality productivity is increasingly apparent. However, we must also be clear that the development and application of AI model technology in China still face challenges such as the lack of high-quality Chinese data resource libraries, insufficient supply of advanced computing power, constraints of basic software and hardware, and an incomplete governance system for AI, among many other issues. There are still numerous bottlenecks, card points, and risk points in accelerating the development of new quality productivity empowered by AGI, which urgently need to be addressed with targeted measures.

(1) Bottlenecks in the Multiplicative Effect of Data Elements

Data is the cornerstone of AI development. China is a major manufacturing country, and the development of the digital economy is in full swing, possessing vast data advantages on both the consumer and business sides. In 2022, China’s data output reached 8.1ZB, a year-on-year increase of 22.7%, accounting for approximately 10.5% of the global data output; in the same year, the scale of China’s data trading market reached 87.68 billion yuan, accounting for about 13.4% globally and about 66.5% in Asia.^[19] However, overall, China’s data production, circulation, and utilization still face problems such as unclear data property rights, non-uniform standards, low sharing levels, and insufficient security guarantees^[21], which directly restrict the full release of the potential of data elements. Firstly, the multi-faceted involvement of subjects related to data elementization makes data confirmation and analysis extremely difficult; coupled with the fact that the market contribution of data varies by application scenario, it is challenging to form a standardized and scalable data value pricing mechanism, leading to obstacles in data resourceization, assetization, and capitalization. Secondly, data trading lacks standardized trading objects and secure trading spaces, resulting in relatively limited in-market data trading volume, making the circulation of data resources inefficient. Thirdly, the fragmentation of government public data departments and the monopoly of social public data platforms have created data barriers and data islands, hindering the progress of cross-subject data openness and sharing, preventing the advantages of basic data resources from being fully converted into advantages for high-quality development.

Furthermore, as mentioned above, the iterative development of AI models directly relies on the sustainable supply of high-quality data; the effective supply of high-quality data comes from the collection, integration, and sharing of multi-source, heterogeneous data. Currently, the technical and institutional barriers that hinder the “activation, movement, and utilization” of data elements in China result in severe issues such as a lack of high-quality data sets in the AI field, an incomplete data supply industry ecosystem, and high costs for enterprises to obtain data resources^[22], which directly restrict the empowering effectiveness of AI models for the rapid development of new quality productivity.

(2) Card Points in the AI Model Technology Ecosystem

1. Bottlenecks in the Effective Supply of Intelligent Computing Power

The history of human social and economic development shows that human power, animal power, steam, and electricity have successively become the main driving forces for the development of productivity in primitive, agricultural, and industrial societies. Entering the digital intelligence society, the importance of computing power is increasingly highlighted due to AI applications. The training, deployment, and inference of AI models require strong support from intelligent computing power^⑤. From the perspective of global trends in the new generation of AI development, advanced AI models often require tens of thousands of card clusters, and the competition for computing resources has become a new focus of competition among digital technology giants. In a sense, the competition of AI models is essentially a competition for computing power. Whoever possesses abundant advanced computing power is likely to win in the competition of large models^[23].

Currently, China’s general AGI research and development is characterized by a “hundred models competing for supremacy,” and the demand for intelligent computing power is surging. The Ministry of Industry and Information Technology’s “Action Plan for the High-Quality Development of Computing Power Infrastructure” estimates that by 2025, China’s overall computing power demand will be about 300 EFLOPS, with a demand for intelligent supercomputing of 105 EFLOPS^⑥. Overall, China is already the second-largest computing power country, but the development of AI computing power in China, especially in terms of computing power suitable for training AI models, still lags significantly behind the global level^[19]. Restricted by foreign high-tech blockades, the domestic production and processing of high-end intelligent chips are significantly behind those in the West, making advanced graphics processing units (GPUs) hard to come by, and the localization of AI intelligent computing power is urgent.

2. Core Algorithms in the AI Field Subject to External Constraints

Practical experience shows that the development and application of AI model technology not only require reliance on intelligent hardware but also depend on the support of a software ecosystem. Among them, the development and iteration of core algorithms fundamentally determine data mining capabilities, which in turn directly affect the practical effectiveness of AI models in empowering new quality productivity. Currently, despite some breakthroughs in certain specialized algorithm fields, overall, most existing domestic AI models are based on foreign open-source deep learning frameworks such as PyTorch or TensorFlow, and the pre-training algorithm frameworks are largely based on Google’s Transformer architecture, resulting in a low level of independent research and development for core algorithms and key software. In other words, compared to developed countries, China has significant shortcomings in the innovation of key common technologies in AI, lacking major original technologies in areas such as analytical reasoning technology, modeling technology, and intelligent computing chips^[24]. It is evident that as major powers engage in increasingly fierce high-tech competition, the “intellectual property barriers” embedded in the AI model technology ecosystem pose real risks of stagnation for China’s AGI development for a considerable time to come.

(3) Risks in the Application of AI Models

Currently, AI models such as Baidu Wenxin, Alibaba Tongyi, Huawei Pangu, Tencent Hunyuan, and SenseTime are emerging in competition, opening up new tracks for China’s intelligent economy. According to the “Research Report on the Map of AI Models in China” released by the New Generation Artificial Intelligence Development Research Center of the Ministry of Science and Technology, the number of AI models developed in China ranks second globally. Indeed, the development of AI technology, like other technological advancements, is a double-edged sword^[25]. Currently, while AI models are unleashing tremendous intelligent productivity, they may also give rise to new issues in digital governance due to the “abuse of large models.”

At present, with the rapid innovation in AGI technology, AI model products or services are emerging one after another. If left unchecked, their rampant generation will inevitably lead to issues such as the proliferation of “digital waste,” the new alienation of digital labor, increasing digital inequality, and escalating data security risks. For promoting the high-quality development of China’s economy, the following two risks brought by the application of AI models need to be approached with caution. First, the risk of “machine replacing humans” losing momentum. The application of AI models is having a disruptive impact on the ways of human material and spiritual production. A significant portion of procedural work will gradually be replaced by intelligent machines. The latest research results from the International Monetary Fund (IMF) indicate that nearly 40% of global employment opportunities are affected by AI^[26]. China is a major employment country and also a major digital economy country. It is evident that the large-scale application of AI models poses practical challenges regarding employment. Second, the compliance risks of AIGC. The low threshold and ease of use of AIGC have given rise to a revolution in knowledge production efficiency, but they also inevitably bring issues such as malicious fabrication of false information and infringement of intellectual property rights. If the “side effects” of AI model technology applications run rampant, they will disrupt the market social environment, impact the order of intellectual property rights, and hinder the surge of innovation and the rapid development of new quality productivity.

3. Practical Approaches to Accelerate the Development of New Quality Productivity Empowered by AI Models

Fostering and accelerating the development of new quality productivity hinges on innovation, with industry as the carrier and potential in the factors^[27]. In the face of the global wave of AGI development, taking the empowering effect of AI models as an opportunity to open up new tracks to promote the emergence of new quality productivity is key to seizing the high ground of development. The new quality productivity represented by tools such as AI models determines the depth and breadth of China’s participation in the Fourth Industrial Revolution. In this context, moving towards a new journey, we must adhere to the important discourse of General Secretary Xi Jinping on new quality productivity as guidance, striving to build a new development pattern that promotes the rapid development of new quality productivity through AI models.

(1) Consolidating the Foundation for AGI Development

Data, algorithms, and computing power are the “three horses” of AI models. Fundamentally, achieving China’s catching up and global leadership in the AGI field requires us to target large models, integrate big data, and lay out large computing power, concentrating efforts in high-quality data element supply, efficient and intensive intelligent computing resource supply, and cultivating an advanced independent AI model technology system to achieve controllability in data, frameworks, and models.

1. Accelerating the Construction of High-Quality Chinese Data Resource Libraries

The lack of high-quality Chinese data resources is a core pain point restricting the independent research and development and application of AI models in the Chinese context. We must leverage the implementation of the “Three-Year Action Plan for Data Elements” (2024-2026) as an opportunity to gather and integrate existing public and social data resources, establish an AI model corpus data alliance oriented towards industries, improve the public data authorization and operation mechanism, promote the collection, access, sharing, processing, and utilization of typical industry data, and strive to activate the existing data resource stock; strengthen the construction of data standards systems, improve data fine labeling levels, and continuously promote the collaborative optimization, reuse enhancement, and innovative integration of data elements. At the same time, actively develop high-quality datasets covering text, images, audio, and video, create synthetic data based on AIGC technology, and cultivate data resource increments, effectively addressing issues of data insufficiency, data noise, and data bias, providing high-quality pre-training datasets and optimized training datasets for AI industry development.

2. Accelerating the Construction of Independent Computing Power Resources and Supply Capabilities

Focusing on a unified national approach, vigorously promote the construction of computing power infrastructure and networks, and address prominent issues such as the uneven distribution of computing power and low utilization rates of computing resources, continuously enhancing the overall supply level of intelligent computing power. The urgent task is to steadily advance the construction of a unified national public computing power service platform, aggregating and coordinating existing computing resources, scientifically laying out the construction of AI intelligent computing centers, forming computing power clusters, efficiently aggregating and allocating multi-cloud computing power, standardizing and prospering the computing power trading algorithm economic ecosystem, and effectively resolving the outstanding issues of “East data, West computing” such as “unable to compute, costly computing, and poor computing”; encourage and support leading domestic public cloud vendors to increase market investment, improve and innovate cloud services, promote the accelerated deployment of state-owned cloud, expand the supply of quality and inclusive computing power, and facilitate small and medium-sized enterprises in realizing low-cost “cloud-based intelligence through data.”

3. Accelerating the Independent Construction of a High-Level AI Model Technology Ecosystem

We must adhere to the integrated design principles of original innovation, integrated innovation, and open innovation, while absorbing international advanced experiences and technologies in algorithms and software, and based on high-level technological self-reliance, leveraging the advantages of a new type of national system to construct an innovation system deeply integrated with government, industry, academia, and research led by enterprises, concentrating efforts on collaborative tackling, and promoting the full-stack technological capability construction of AI models from “chips—frameworks—models—applications,” breaking through card points in critical core technology systems, and maintaining leading advantages in foundational frontier fields such as quantum computing prototypes, brain-like computing chips, and carbon-based integrated computers, while striving for new breakthroughs in new architectures, new algorithms, performance evaluation, and foundational software and hardware support.

(2) Creating New Highlands for AI Industry Development

In July 2017, the State Council issued the “New Generation Artificial Intelligence Development Plan,” proposing the acceleration of cultivating AI industries with significant leading and driving roles, promoting the deep integration of AI with various industries, and forming a smart economy model driven by data, human-machine collaboration, cross-border integration, and co-creation and sharing. The report of the 20th National Congress of the Communist Party emphasized accelerating the development of the digital economy, promoting the deep integration of the digital economy with the real economy, and creating internationally competitive digital industry clusters^[28]. In December 2023, the Central Economic Work Conference pointed out that technological innovation should drive industrial innovation, especially using disruptive and cutting-edge technologies to give birth to new industries, new models, and new driving forces, and develop new quality productivity^[29]. Currently, AI model technology is leading a new wave of digital technology revolution, promoting the application of AGI technology innovation scenarios, and creating new highlands for the development of the AI model industry, which is a strategic measure to accelerate the development of new quality productivity and promote new industrialization, enhancing the modernization level of China’s industrial system.

According to data estimates released by the China Academy of Information and Communications Technology, in 2022, the scale of China’s core AI industry reached 508 billion yuan, a year-on-year increase of 18%^[30]. During the 14th Five-Year Plan period, we must align with the trend of intelligent economy development, coordinating the landing and promotion of AI model technology from both the industrialization of AI and the AIization of industries. On one hand, we should promote the core intelligent economy industries to strengthen and stabilize the supply chain, introducing AI model technology applications in the foundational software field, comprehensively improving the performance of domestic software in office, design, editing, etc., and enriching the categories of independent innovation software; we should promote the building and extending of AIGC industries, creating new digital cultural innovation business formats such as text generation, image generation, audio and video generation, cross-modal generation, 3D asset generation, virtual human generation, and game strategy generation, enhancing the continuity and competitiveness of digital content industry development. On the other hand, we should actively develop vertical models targeted at specific industries, focusing on exploring applications in fields such as healthcare, e-commerce, finance, robotics, and autonomous driving, promoting more scenarios and industries of “AIGC +” innovative development, and cultivating new competitive advantages in the intelligent economy industry.

(3) Promoting High-Quality Development of Digital Labor

Employment is the most fundamental aspect of people’s livelihood. With the development of AGI, digital labor faces new opportunities brought by “AI empowerment” while also encountering a new wave of unemployment shocks from “machine replacing humans.” Supporting and regulating the development of new employment forms, strengthening the protection of rights and interests of flexible employment and new employment forms, urgently requires leveraging the system advantages of the socialist market economy, focusing on making solid progress in promoting high-quality development of digital labor through the application of AI model technology.

Firstly, we should support and regulate new occupations in digital labor. The application of AI model technology has greatly expanded the new space for intelligent economy, leading to the emergence of new occupations such as AI trainers, prompt engineers, and algorithm engineers. Government regulatory departments must plan ahead and take the lead in providing strong institutional guarantees in terms of occupational norms, skills training, and career development. Secondly, we should emphasize and guide the transformation and upgrading of crowdsourced micro-labor. For crowdsourced micro-laborers, the employment shocks brought by the application of AI model technology are the most direct. Taking the data annotation industry as an example, general data annotation will be replaced by AI. However, it should also be noted that as the vertical application of large models accelerates, the market demand for industry-specific data is also increasing. For a considerable period, the multi-modal datasets adapted to the industry will still require professional personnel for annotation, naturally giving rise to new positions for detailed annotation. In this regard, government-enterprise cooperation can be strengthened, and training in professional knowledge and new skills can be provided to help existing data annotation practitioners enhance their skills and transition to new jobs. Thirdly, we should promote and improve the work experience of on-demand labor. In the development of the gig economy, urging platform companies to adopt benevolent algorithms and promote algorithm fairness is fundamental to improving the working conditions and labor conditions of gig workers. The application of AI model technology reconstructs platform algorithm systems, objectively providing new opportunities for improving the management of on-demand labor algorithms. We should encourage and guide leading platform companies to innovate digital labor through AI empowerment, focusing on areas such as task assignment, service pricing, digital reputation management, and customer communication, enriching interface operation options, enhancing algorithm management transparency, and strengthening the proactive experience of digital labor.

(4) Improving the Governance Ecosystem of AI Models

Law establishes the foundation, and customs take shape from the bottom. In the current and forthcoming period, we should take the implementation of the “Management Measures for Generative Artificial Intelligence Services” as an opportunity to build and improve an AI governance system that balances innovative development with safety and trustworthiness, and combines government regulation with industry self-discipline and autonomy, continuously promoting the healthy development and standardized application of AI models. On one hand, we should promote inclusive and prudent, normalized regulation of AGI. For AIGC that provides services to the public, we should focus on aspects such as foundational algorithm design, training data source selection, manual annotation rules, model generation and optimization, content safety, and service fairness, setting up AI guardrails, strengthening review and evaluation, and providing compliance guidance. On the other hand, we should promote industry self-discipline and autonomy for AI models. In the short term, platform companies must fulfill their responsibilities as AIGC producers, actively exploring best practices in respecting intellectual property rights and business ethics, protecting user information and privacy, value alignment, and preventing excessive reliance on or malicious use of generated content. In the medium to long term, leading platform companies should actively engage in building responsible AI models, exploring core issues such as addressing the crisis of human subjectivity, implementing technological ethical requirements, and promoting sustainable human development, contributing Chinese solutions and wisdom to the construction and improvement of the global governance system for AI models.

Notes:

① New quality productivity is generated by revolutionary breakthroughs in technology, innovative configurations of production factors, and deep transformation and upgrading of industries, characterized by innovation, with quality as the key and advanced productivity as the essence.

② Meta’s ESMFold model can predict protein structures and sequences based on input sequences, completing predictions for over 600 million rare substances in just two weeks. Huawei’s Pangu drug molecule model can achieve efficient optimization of drug molecules based on graph-structured drug input, generating 100 million drug molecules with a novelty rate of 99.68%.

③ In July 2023, a deep intelligence open-source multi-intelligent framework called MetaGPT was launched, which helps users build their own virtual companies with intelligent agents as employees.

④ On January 11, 2024, OpenAI’s GPT Store officially launched, and within just two months, over 3 million GPTs had been created on the platform.

⑤ Data shows that in recent years, the parameter scale of AI models has grown at an average annual rate of 400%, and the demand for AI computing power has increased by over 150,000 times, far exceeding Moore’s Law.

⑥ 1 EFLOP of computing power is equivalent to 100 quintillion calculations per second.

References:

[1] Zhou Hongyi. AI Models: Driving Forces of the New Industrial Revolution and New Opportunities for China’s Development [J]. New Economic Guide, 2023(3): 41-46.

[2] China Academy of Information and Communications Technology. Blue Paper on the Governance of Large Models: From Rules to Practice [EB/OL]. (2023-11-28) [2023-12-17].

https://www.doc88.com/p-99329852318414.html.

[3] Chen Yongwei. GPT as GPT: Opportunities and Challenges of the New Generation of Artificial Intelligence from the Perspective of General Purpose Technology [J]. Reform, 2023(6): 41-58.

[4] Gao Fan. The Logic, Multidimensional Connotation, and Era Significance of the Proposal of “New Quality Productivity” [J]. Review of Political Economy, 2023(6): 127-145.

[5] Compilation Bureau of the Central Committee of the Communist Party of China. Selected Works of Marx and Engels: Volume 2 [M]. Beijing: People’s Publishing House, 2012: 453.

[6] Accenture. Generative Artificial Intelligence: The New Era Accessible to Everyone [EB/OL]. (2023-05-29) [2023-12-17]. https://www.doc88.com/p-30687588813721.html.

[7] China Academy of Information and Communications Technology, JD Research Institute. White Paper on Artificial Intelligence Generated Content (AIGC): 2022 [EB/OL]. (2023-04-19) [2023-12-17].

http://www.199it.com/archives/1586707.html.

[8] Accenture. Technology Outlook 2023: When Atoms Meet Bits, Building a New Foundation for Digital-Physical Integration [EB/OL]. (2023-07-17) [2023-12-17].

https://max.book118.com/html/2023/0717/5230104241010243.shtm.

[9] Chen Jin. Driving New Quality Productivity with New Technologies and New Industries [J]. Economy, 2024(1): 22-24.

[10] Central Committee of the Communist Party of China. Excerpts from Xi Jinping’s Discourses on Strengthening Cyber Power [M]. Beijing: Central Literature Publishing House, 2021: 121.

[11] Compilation Bureau of the Central Committee of the Communist Party of China. Collected Works of Marx and Engels: Volume 5 [M]. Beijing: People’s Publishing House, 2009: 483.

[12] Tao Feng. Agency, Simulation, and Technology: A Philosophical Interpretation of AI Literary Production [J]. Philosophical Research, 2023(3): 56-67.

[13] Sen Jian, Nichi Hiroshi. Digital Capitalism [M]. Nomura Research Institute (Dalian) Technology Co., Ltd., Translated. Shanghai: Fudan University Press, 2020: 131.

[14] 2024 AIGC Application Layer Top Ten Trends [EB/OL]. (2024-01-05) [2024-01-17]. https://www.163.com/dy/article/INLCVFLS051998SC.html.

[15] Ouyang Rihui, Liu Yuhong. Theoretical Logic and Realization Path of Generative Artificial Intelligence (AIGC) in Manufacturing Industry [J/OL]. Journal of Xinjiang Normal University (Philosophy and Social Sciences Edition): 1-14 [2024-02-17]. https://doi.org/10.14100/j.cnki.65-1039/g4.20240016.001.

[16] Peng Lan. AIGC and New Survival Characteristics in the Intelligent Era [J]. Nanjing Social Sciences, 2023(5): 104-111.

[17] PENG S, KALLIAMVAKOU E, CIHON P, et al. The Impact of AI on Developer Productivity: Evidence from GitHub Copilot [EB/OL]. [2023-12-10].

https://www.microsoft.com/en-us/research/publication/the-impact-of-ai-on-developer-productivity-evidence-from-github-copilot/.

[18] Qingyuan. Gartner Report Predicts: By 2026, More than 100 Million People Will Work with Generative AI [EB/OL]. (2023-10-23) [2023-12-20].

https://www.ithome.com/0/697/699.htm.

[19] 2023 Digital China Annual Report [EB/OL]. (2024-01-16) [2024-01-20]. https://www.doc88.com/p-77587671986915.html.

[20] 2024 Annual Report on Five Major Mobile Application Market Predictions [EB/OL]. (2023-12-17) [2023-12-19].

https://www.sohu.com/a/744593340_121238562.

[21] Shi Jianxun. The Major Strategic Significance and Realization Path of Accelerating the Formation of New Quality Productivity [J]. Research on Financial Issues, 2024(1): 3-12.

[22] China Academy of Information and Communications Technology. White Paper on Data Elements: 2023 [EB/OL]. (2023-11-09) [2024-01-20]. https://www.doc88.com/p-99539155938092.html.

[23] Huang Zaisheng. New Domains of Value Movement in Generative Artificial Intelligence and Digital Capitalism [J]. Foreign Theory Dynamics, 2024(1).

[24] Wang Shuixing, Liu Yong. Intelligent Productivity: A New Quality of Productivity [J]. Contemporary Economic Research, 2024(1): 36-45.

[25] Compilation Bureau of the Central Committee of the Communist Party of China, Office of the Leading Group for the Thematic Education of Xi Jinping’s New Era of Socialism with Chinese Characteristics. Excerpts from Xi Jinping’s New Era of Socialism with Chinese Characteristics [M]. Beijing: Central Literature Publishing House, 2023: 418.

[26] CAZZANIGA M, JAUMOTTE F, LI L, et al. Gen-AI: Artificial Intelligence and the Future of Work Prepared [R]. Washington DC: International Monetary Fund, 2024.

[27] Yu Donghua, Ma Luming. New Quality Productivity and New Industrialization: Theoretical Interpretation and Interactive Paths [J]. Tianjin Social Sciences, 2023(6): 90-102.

[28] Xi Jinping. Holding High the Great Banner of Socialism with Chinese Characteristics to Unite and Strive for the Comprehensive Construction of a Modern Socialist Country: Report at the 20th National Congress of the Communist Party of China [N]. People’s Daily, 2023-10-26(1).

[29] The Central Economic Work Conference Held in Beijing [N]. People’s Daily, 2023-12-13(1).

[30] Gu Yekai. Accelerating the Application of Artificial Intelligence Industry [N]. People’s Daily, 2023-03-25(5).