The Integration of Generative Artificial Intelligence (AIGC) into Manufacturing Industry
Theoretical Logic and Implementation Path
Abstract
Entering the Web 3.0 era, the new generation of artificial intelligence technology represented by AIGC has brought significant impacts on the development of traditional manufacturing industry. AIGC has basic characteristics of general-purpose technology such as penetration, substitutability, collaboration, and creativity, along with key technical-economic characteristics including adaptability, operability, and parallelism, which are of great significance for improving the micro-operational efficiency of the manufacturing industry and promoting macro high-quality development. From a macro perspective, AIGC relies on scale effects and innovation-driven strategies to accelerate the release of economic potential while advancing the transformation of the intelligent manufacturing industry; from a micro perspective, AIGC mainly optimizes resource allocation efficiency by leveraging operational effects and achieves an increase in total factor productivity through the flywheel effect. However, there are still many constraints in the AIGC industry chain that hinder the advancement of intelligent transformation in the manufacturing industry. Based on this, the text discusses the theoretical logic and implementation paths for the integration of AIGC into the manufacturing industry from the foundational, operational, and application layers.
Keywords
AIGC; Intelligent Manufacturing; Digital Economy; Technical-Economic Characteristics
Author Introduction
Ouyang Rihui Vice Dean of the China Internet Economy Research Institute at Central University of Finance and Economics, Researcher, Doctoral Supervisor
Liu Yuhong Assistant Researcher at the Research Department of the China Marketing Association
1. Introduction of the Problem
With the rapid development of the digital economy, the emerging artificial intelligence-generated content (AIGC) is gradually replacing professional generated content (PGC) and user-generated content (UGC) in some fields, not only promoting the upgrade of traditional industries’ production methods, circulation paths, and sales models but also driving the digital and intelligent transformation of industries. Currently, artificial intelligence is a key node for high-quality development in the digital economy, and AIGC has become the most characteristic form of production in this era. On November 30, 2022, OpenAI released the representative chatbot model of AIGC, ChatGPT, which gathered one million users within five days, quickly breaking the record of traditional social media platform Meta. According to Similarweb statistics, in April 2023, ChatGPT’s monthly visits reached 1.76 billion, providing convenience to users through a new form of content generation, from basic robotic chatting to highly interactive customer service assistance. At this stage, AIGC is accelerating its integration into the economy and society, attracting widespread attention from the government and academia.
In July 2017, the State Council issued the “New Generation Artificial Intelligence Development Plan”, proposing to accelerate the cultivation of artificial intelligence industries with significant leading roles, promote the deep integration of artificial intelligence with various industries, and form a data-driven, human-machine collaboration, cross-border integration, and co-creation sharing intelligent economic form. Based on data elements, and supported by digital technologies such as artificial intelligence and the Internet, driving the digitalization and intelligent transformation of traditional industries has become the core of high-quality development in the digital economy. In 2019, President Xi Jinping further pointed out in his congratulatory letter to the Third World Intelligence Conference: “China attaches great importance to innovative development, taking the new generation of artificial intelligence as a driving force to promote leapfrog development in science and technology, optimize industrial upgrading, and elevate overall productivity to achieve high-quality development.” On February 6, 2023, the Central Committee of the Communist Party of China and the State Council issued the “Quality Power Country Construction Outline”, emphasizing the need to accelerate the deep application of new technologies such as big data, networks, and artificial intelligence, and promote the integrated development of emerging technologies and advanced manufacturing industries. Currently, the position of artificial intelligence in the national economy is becoming increasingly prominent, and AIGC, as an important branch of the new generation of artificial intelligence, has attracted significant attention. How to promote the deep integration of artificial intelligence with traditional manufacturing has become a focal issue for the Party and the country.
Research on the integration and transformation of AIGC and industries in academia is mainly concentrated in journalism, publishing, media, and education, while relevant research in the manufacturing sector is relatively scarce. Existing literature has confirmed the catalytic role of AIGC in the intelligent transformation of real industries from different perspectives, but the mechanism of AIGC’s role in the emerging intelligent manufacturing industry remains unclear, failing to meet the strategic expectations of the Party and the country for high-quality development in manufacturing. Against this backdrop, what kind of impact will AIGC bring to the manufacturing industry? Through what paths can AIGC deeply integrate into manufacturing? The author intends to explore the theoretical mechanisms and constraints of AIGC’s integration into the manufacturing industry, summarizing the implementation paths of AIGC’s incorporation into manufacturing based on various applications and explorations of AIGC.
2. The Connotation and Technical-Economic Characteristics of AIGC
(1) Analysis and Connotation of AIGC Concept
AIGC is a typical generative model in the field of machine learning, distinguished from discriminative models by its conditional probability modeling, possessing the main characteristics of class-oriented and class-based joint probability modeling. The “Artificial Intelligence Generated Content (AIGC) White Paper” released by the China Academy of Information and Communications Technology and JD Exploration Research Institute defines AIGC as a special set with dual attributes of content characteristics and technical characteristics; it can be viewed as a content, a content production method, or a category of technology. Specifically, AIGC is a specific type of network information resource, a product of intelligent synthesis of a series of content such as text, images, and audio achieved through generative machine learning models, an abstract carrier of highly condensed big data, and an important means for data elements to complete integration, processing, output, and other processes to realize value addition.
From the perspective of the evolution of Internet forms, AIGC is a new form of content generation following PGC and UGC, achieving a new dimension of data element enrichment and utilization (see Table 1).
In the traditional Web 1.0 era, PGC was the main form of Internet content generation, with data sourced from specialized databases in specific fields, characterized by professionalism, verticality, and authority. Experts with specialized knowledge and technology were the main data interaction group in this model, for example, Dingxiang Doctor. At this stage, the application mining of data elements was in its early stage, processing data through traditional web technologies such as TF-IDF and LDA, with relatively low interactivity and fluidity. In the UGC-dominated Web 2.0 era, the subjects of data interaction shifted from expert groups in various fields to a broad base of Internet users, with widespread and fragmented individual data providers becoming the basic data source, and universal, personalized, and diversified data becoming the significant hallmark of this form. Data elements were further developed, and various algorithm technologies such as XPath, CSS Selector, and Hadoop enriched users’ communication experiences on the Internet. With the continuous development of the new generation of artificial intelligence technology, the arrival of AIGC gradually pushed the Internet form into the Web 3.0 era. Integrating Natural Language Processing (NLP) and Natural Language Generation (NLG) methods, AIGC has completely changed the traditional content generation model, enabling AI machine models to automatically and intelligently provide innovative content and services based on user needs, such as ChatGPT. The subjects of data interaction shifted from different user groups to a new type of “human-machine interaction”. Technologies such as convolutional neural networks and Transformer large models enabled the parameters of deep learning models to achieve leaps in upgrades, and the integration of massive datasets made the content generated by models more feasible and operable, fully releasing the value of data elements. On this basis, the data in the AIGC model becomes more innovative, fluid, and interactive. AIGC has a tremendous impact on existing production methods, expected to change traditional industry models and development paradigms, becoming an important driving force for promoting new social forms of “human-machine symbiosis” and “human-machine co-creation”.
(2) The Industrial Structure Levels of AIGC
With the further development of artificial intelligence technology and the digital economy, the industrial chain structure of AIGC has begun to take shape, which can be specifically divided into foundational, operational, and application layers (see Figure 1).
The foundational layer is the foundation of the AIGC industry. The development of AIGC relies on the support of data and computing power, where data is the basic “fuel” for AIGC, and computing power is the guarantee for AIGC’s operation, together forming the main components of the foundational layer. Data integration is a prerequisite for AIGC to generate, store, analyze, and apply data, as well as the foundation for the efficient operation of the entire industrial chain. Before the maturity of zero-shot learning (ZSL), foundational databases, big data platforms, and data centers remain important sources of massive data in the AIGC industry. The foundation of computing power is crucial for AIGC to successfully convert the value of integrated data into effective output. The configuration of foundational hardware such as smart chips, smart sensors, and cloud computing platforms is an indispensable part of the industrial chain, determining the “IQ” of artificial intelligence.
The operational layer is the core driving force of AIGC, with algorithm models providing technical support for promoting the cross-border integration of AIGC with various industries. First, there are general-purpose technologies. In AIGC, general-purpose technologies refer to a series of systematic technologies that build intelligent agents capable of general task-solving and continuous autonomous learning, possessing perception, cognition, decision-making, and planning capabilities, thus endowing them with human-like intelligent characteristics and levels. For example, machine learning and knowledge graphs are important supports at the technical layer. Second, there are key technologies, mainly including Variational Autoencoders (VAE), Generative Adversarial Networks (GAN), Transformer models, and Diffusion models. Based on data and computing power, through continuous optimization and innovation of key algorithm technologies, AIGC can generate text, images, audio, video, and other content that is similar to or even surpasses human creative levels, bringing new changes and breakthroughs to economic and social development. Third, there are pre-trained large models. Pre-trained large models are a special auxiliary technology in the development of artificial intelligence, capable of training on large amounts of unlabeled data to learn the underlying rules and features within the data. Previously, the content and quality of various generative models could not meet the demands of different scenarios. Pre-trained large models in computer vision (CV), natural language processing (NLP), and multimodal pre-trained large models have significantly improved the generalization ability of AIGC, enabling it to make relatively accurate predictions when facing new data. Therefore, driven by various technologies at the operational layer, AIGC can greatly optimize the processing efficiency of various links in the industrial chain, becoming a key support for driving cross-border integration of industries. The application layer focuses on the connection between algorithm models and user demand scenarios, which is key to the realization of intelligent industry by AIGC. Driven by the foundational and operational layers, the application layer applies various text, image, audio, and video generation content to production, circulation, sales, and management in other industrial chain links, making the emergence of new intelligent industry transformations by AIGC possible. In this process, based on different value creation logic, the generated content of AIGC can be divided into four categories: production of directly consumable content, production of high-value-added content combined with underlying systems, provision of content production auxiliary tools, and provision of systematic solutions. From the production link perspective, AIGC completes product reshaping by directly participating in product production, assisting in product production, and adding value to products, inducing the formation of high-quality and high-efficiency production paradigms. From the circulation link perspective, the application of AIGC mainly lies in the intelligence and automation of circulation equipment, promoting the promotion of intelligent transportation and automated delivery, thereby accelerating the establishment of a digital transportation system. From the sales link perspective, the integrated data based on various pre-trained large models can make more precise predictions and judgments when responding to different consumer demands and preferences, enhancing the operability and practical significance of targeted sales and user mining. From the management link perspective, solutions and decision-making involving AIGC can avoid human planning errors to some extent, and systematic generative decisions can provide feasible reference norms for the direction of enterprise development.
In summary, data, computing power, algorithm models, and scenario applications are the key elements of the AIGC industrial chain, gradually forming a three-layer structure with data and computing power at the core of the foundational layer, algorithm models at the core of the operational layer, and scenario applications at the core of the application layer. Therefore, solidifying the foundational layer’s industrial base, optimizing the operational layer’s algorithm models, and improving the application layer’s scenario matching become the main paths for driving high-quality development in the AIGC industry.
(3) The Technical-Economic Characteristics of AIGC
As a new generation of artificial intelligence, AIGC possesses the basic characteristics of general artificial intelligence, namely, industry penetration (Pervasiveness), factor substitutability (Substitution), production synergy (Synergy/Cooperativeness), and content creativity (Creativeness). These four points are the technical-economic characteristics commonly possessed by General Purpose Technology (GPT) and are also the technical basis for triggering a new technical-economic paradigm shift following the previous round of industrial technological revolution. Self-adaptability (Self-adaptation), operability (Operability), and parallelism (Parallelism) are fundamental distinctions between AIGC and other technologies in the GPT field, and they are significant indicators of the transition from artificial intelligence 1.0 to 2.0.
Self-adaptability is the most basic and prominent technical-economic characteristic of AIGC, allowing it to autonomously adjust its generative planning and decision-making tendencies based on its perception of the environment and the system itself, thus adapting to various dynamic environments and uncertain scenario demands, leading to better flexibility and reliability. Taking ChatGPT as an example, the generated content of this application is not fixed or templated; AIGC can make more targeted choices and adaptations based on different users and actual scenario demands. This is mainly attributed to the massive data integration and pre-trained large models behind AIGC, where the model learns knowledge from large-scale unsupervised data across various fields and transfers that knowledge to specific tasks based on actual conditions, not only omitting repetitive training and data labeling processes but also greatly enhancing the ability for generalized applications. In recent years, the capability of data integration in the AIGC field has rapidly developed. In 2018, Google proposed BERT, with a model parameter count of 300 million; in February 2019, OpenAI launched GPT-2 with 1.5 billion parameters; in June 2020, OpenAI’s GPT-3 surpassed 175 billion parameters; and in March 2023, OpenAI released GPT-4 with a parameter count of 1.8 trillion. The massive data means a broader matching space and a more comprehensive generation range, and the value of data elements is fully realized under the operation of AIGC. “Data empowerment – scenario self-adaptation” has become one of the typical characteristics of AIGC.
Operability refers to the characteristic of intelligent operation of the system, which is the second prominent technical-economic characteristic of AIGC. The foundational layer’s data and computing power are prerequisites for operability, while the algorithm models and key technologies of the operational layer are crucial for achieving intelligent operation. In 2021, OpenAI released the multimodal model (Contrastive Language-Image PreTraining, CLIP), which can complete feature extraction and comparison of different texts and images while using ultra-large-scale text-image training, and realize mutual understanding across modalities through similarity calculations, marking the development of AIGC entering a new stage of multimodal models. Based on data and computing power, driven by core algorithm models, machines gradually produce “wisdom” capable of “independently scheduling and allocating resources within adjustable limits of the system”, which is the origin of AIGC’s operability. Specifically, AIGC’s operability manifests in two aspects: first, the intelligent operation in terms of factor allocation. The essence of AIGC’s operation is the allocation process of different factors, where “artificial intelligence” replaces traditional universal labor, with algorithm models centered on data elements occupying a dominant position in the operation. AIGC synchronously adjusts the allocation of resources in different stages of operational scenarios, where factors such as land, labor, capital, technology, and data may become key drivers of production at specific links. This process of “discovery” and coordinated allocation is the core of operability. Second, intelligent operation in terms of industrial applications. Under the influence of industry penetration, the characteristics of AIGC’s intelligent operation will subtly integrate into various links of other industries, accelerating the formation of intelligent industrial chains. For example, in the intelligent education industry driven by AIGC, artificial intelligence technology can quickly and conveniently visualize abstract knowledge that traditional education models find difficult to express, providing students with real-time and accurate feedback on their learning. The impact of AIGC will empower changes in teaching relationships and educational models, further stimulating innovations in new educational paradigms. In short, “model empowerment – intelligent operation” is the main expression of operability. Parallelism is the third important technical-economic characteristic of AIGC, referring to the ability of computer systems to perform calculations or operations simultaneously and complete two or more tasks at the same time. In 2017, Uber released the Horovod data parallel training framework, attempting to solve the scalability issues of parameter aggregation communication in traditional parameter server distributed training architectures by borrowing the Ring AllReduce technology from high-performance computing, marking the beginning of the integrated development of artificial intelligence and high-performance computing technologies. Subsequently, significant advancements were achieved in data parallelism, model parallelism, and other fields of artificial intelligence technology, making AIGC’s parallelism advantages increasingly prominent. On one hand, AIGC operates without restrictions of time, location, or language, capable of running in different environments and scenarios, processing multiple tasks or data streams simultaneously. This means that AIGC can exhibit super high efficiency, several times that of single-task processing flows, within the same time frame, enhancing the operational efficiency of economic systems. On the other hand, the promotion and application of parallel engineering centered around AIGC will further drive the transformation and evolution of production methods and organizational forms in traditional production industries, represented by manufacturing. “Spatiotemporal empowerment – multiple parallel tasks” will become a fundamental characteristic of the AIGC industry in the future.
3. The Mechanism of AIGC’s Influence on the Development of Manufacturing Industry
In June 2023, the Ministry of Industry and Information Technology and four other departments jointly issued the “Implementation Opinions on Improving the Reliability of Manufacturing Industry” (hereinafter referred to as “Opinions”), clearly stating that the development of the manufacturing industry should “deepen the application of digital technology in improving reliability”. The “Opinions” proposed to promote the application of digital technology throughout the entire process of product demand analysis, design and development, manufacturing, inspection and testing, and maintenance and guarantee, using model-based systems engineering, digital twins, reliability simulation, and other technologies to improve product reliability design levels, promote the digital transformation of manufacturing equipment, and facilitate the deep application of technologies such as sensing, machine vision, automation control, and advanced measurement in manufacturing processes. At present, digital technology occupies an important position in the high-quality development of the manufacturing industry, and AIGC has become a key factor influencing the development and transformation of traditional manufacturing in the digital age. The author will analyze the deep mechanism and transmission paths of AIGC’s influence on the development of the manufacturing industry, focusing on issues related to improving economic operational efficiency, promoting total factor productivity growth, accelerating the release of economic potential, and generating social welfare and losses based on principles of microeconomics, welfare economics, and related theories.
(1) Mechanism for Promoting Macro High-Quality Development
The three key technical-economic characteristics of AIGC, namely self-adaptability, operability, and parallelism, amplify the improvement of manufacturing operational efficiency at the micro level to a sufficient extent at the macro level, laying a foundation for high-quality development of the manufacturing industry through operational effects and flywheel effects. At the macro level, the scale effects and innovation-driven strategies of AIGC become important paths for improving macro total factor productivity, accelerating the release of economic potential, and promoting the transformation of intelligent manufacturing.
-
Promoting Scale Effects and Accelerating Economic Potential Release
Generally speaking, the scale expansion of various links in the manufacturing industry chain often accompanies various cost obstacles, and avoiding scale inefficiencies is a key issue for enterprise expansion. AIGC has characteristics such as self-adaptability, operability, and parallelism, and the massive and large-scale integration of data is the foundation of its high-efficiency operation, making the development and evolution of AIGC inherently inclined towards “economies of scale”. First, the application of AIGC enables the digitalization and virtualization of some business operations in traditional industries, allowing some business volumes and products to exert scale effects without being restricted by time and space, thereby expanding the production scale of the industry itself. Second, the application of AIGC has explored potential demands beyond the original market, further promoting the expansion of production scale.
On one hand, the innovation generated by AIGC in the development of the manufacturing industry creates corresponding market demands; on the other hand, the significant improvement in production efficiency brought by AIGC stimulates potential market demand, subtly generating scale effects during the integration of AIGC into the manufacturing industry. Finally, appropriate policy guidance can promote the formation of economies of scale, for example, the government can support AIGC projects that align with sustainable development to enable enterprises to achieve adaptive expansion of production scale.
Based on this, the expansion of foundational layer data sets and the large-scale construction of computing power infrastructure can directly enhance AIGC’s operational capacity and breadth. The large-scale application of algorithm technologies and models at the operational layer can comprehensively improve AIGC’s feasibility for personalized responses in different scenarios, while the integration and large-scale promotion of AIGC across various stages of the application layer can effectively enhance total factor productivity. In other words, the technical-economic characteristics of AIGC enable it to realize economies of scale in a wider range and more scenarios, thereby amplifying various positive impacts at the micro level and further releasing economic potential.
2. Elucidating Innovation-Driven Strategies to Promote Intelligent Industry Transformation
The flywheel effect of AIGC, through the continuous expansion of micro-application scenarios, is expected to sustain the operation of the value creation flywheel over a longer period. In the long term, innovation will stimulate more, faster, and broader innovations in this flywheel, driving the transformation of intelligent industries and macro high-quality growth. This can be divided into four stages: the first stage is the initial integration of AIGC into various processes of the manufacturing industry, resulting in a series of intelligent transformations that significantly enhance operational and production efficiency. The second stage sees the emerging products, production methods, operational models, and organizational structures generated by the integration of AIGC into manufacturing radiating globally, helping to capture the opportunities brought by AIGC. The third stage prompts individuals, enterprises, and governments to reassess the role of AIGC, encouraging them to adapt and explore the value of AIGC. The fourth stage involves increasing investments of human and material resources into AIGC-driven innovation, with emerging innovative products igniting a new round of innovation flywheels, forming a virtuous cycle and achieving high-quality development in intelligent manufacturing.
Once the flywheel of AIGC starts turning, each stage may produce various new combinations of products, methods, organizations, and technologies. Innovation-driven strategies become AIGC’s core capability, shifting application scenarios from the past of “blue-collar + repetitive labor (quality inspection, customer service, etc.)” to “white-collar + knowledge innovation” application fields, bringing greater value space and more flywheel effects. However, potential limitations exist at each layer of the innovation flywheel, which may slow down the progress of intelligent transformation. For example, the pre-trained large models at the technical layer may not meet expectations in the process of expansion and improvement. However, AIGC has greatly expanded the development routes of technological innovation, making it possible to predict, avoid, and overcome these obstacles, thereby achieving macro high-quality development.
From a macro perspective, due to the characteristics of AIGC such as self-adaptability, operability, and parallelism, the “data + computing power + algorithms” driven “model prediction” becomes key to various new models, new business forms, and new services, significantly improving total factor productivity and alleviating market failures caused by information asymmetry to some extent, enabling more precise matching between supply and demand. From a micro perspective, the intelligent industrial chain incorporating AIGC helps achieve higher expected goals in processes such as research and development design, production control, customer operation, and business sales. On the demand side, intelligent and personalized products and services help enhance consumption quality, improve the consumption environment, and adjust consumption structure; on the supply side, efficient and adaptable operational mechanisms help adjust industrial structure, optimize factor allocation, and enhance product quality. Therefore, under the dual effects of supply and demand, it helps further release economic potential. A McKinsey report predicts that AIGC could bring $4.4 trillion in annual benefits to the global economy—approximately the GDP of Germany, the fourth largest economy in the world. In the future, the new flywheel will drive the continuous high-quality development of the manufacturing industry.
(2) Mechanism for Improving Micro Operational Efficiency
The core role of AIGC in integrating into the manufacturing industry is to leverage foundational computing power and algorithm models to fully utilize data elements that carry special value, generating one or multiple measurable outcomes through large-scale computational processing and predictive matching in different scenarios, addressing specific business challenges. In this process, due to the presence of self-adaptability and operability, AIGC can utilize large-scale data analysis and mining from the foundational layer to correct information incompleteness at various links in the industrial chain, gradually promoting the adaptation of algorithm models at the operational layer to scenarios, thereby improving operational efficiency in the application layer and optimizing the allocation of production factors to enhance overall social welfare. The parallelism and creativity of AIGC drive the intelligent integration of AIGC and the manufacturing industry, helping cultivate high-end production factors, innovate production modes across industries, enhance productivity and production quality, and ultimately promote the improvement of total factor productivity. At present, the application paradigm of artificial intelligence in the manufacturing industry has begun to take shape, and the exploration of AIGC is gradually covering various links of the manufacturing industry, including research and development design, production control, customer operation, and business sales. From both demand and supply perspectives, combined with the technical-economic characteristics of AIGC, the micro-mechanisms of AIGC can be divided into two main pathways (see Figure 2).
-
Leveraging Operational Effects to Optimize Resource Allocation Efficiency
The operational effects of AIGC refer to the ability of AIGC to fully leverage its “intelligent operation” characteristics to improve the compatibility and application efficiency of various processes in the manufacturing industry chain, gradually achieving specific goals and tasks through learning and flexible application of data, significantly reducing friction costs and organizational costs in economic operations, and ultimately enhancing micro-operational efficiency through optimized resource allocation.
In the research and design phase, AIGC can enhance early research analyses regarding market reports, ideas, and drafting of products or solutions, utilizing “virtual design” and “virtual simulation” technologies to accelerate intelligent operations in the research and development phase. AIGC can directly enhance product design, helping product designers more effectively choose and utilize materials, while also optimizing manufacturing design and reducing various logistics and production costs. Additionally, AIGC can improve product testing processes and product quality, shortening testing times for complex systems by drafting scenarios and describing test candidates, thereby producing higher-quality products and enhancing their market appeal.
In the production control phase, AIGC will promote the integration of software engineering and traditional production, driving the intelligent transformation of manufacturing production control. First, AIGC can assist software engineers and product managers in analyzing, cleaning, and labeling large amounts of data, such as user feedback, market trends, and existing system logs, promptly focusing on and guiding customer demands to optimize production planning. Second, engineers can utilize AIGC to create multiple IT architecture designs and iterate potential configurations, accelerating system design and shortening time to market for products. Third, AIGC can help engineers refine coding work by assisting in drafting, quickly finding prompts, and serving as an easily browsed knowledge base, reducing development time. Lastly, based on insights from system production logs, user feedback, and performance data, AIGC can help diagnose problems and provide corrective suggestions.
In the customer operation phase, intelligent services centered around digital self-service can improve customer experience and operational pressure. The new customer service model integrated with AIGC will fundamentally transform traditional customer operation functions. First, AIGC’s customer self-service can respond instantly and personalized, automatically replying to more customer inquiries by improving the quality and efficiency of automated channel interactions. Second, the efficient “intelligent operation” characteristics of AIGC enable it to promptly retrieve specific data held by enterprises when addressing customer-reported issues, responding based on actual scenarios, thereby effectively increasing the resolution rate of initial issues. Third, AIGC can provide customers with effective suggestions and assistance in real-time, significantly shortening response times for customer operations.
In the business sales phase, AIGC’s main driving force is based on text communication and large-scale personalization capabilities, which can create personalized information according to customers’ personal interests, preferences, and behaviors, as well as complete tasks such as drafting brand advertisements, titles, slogans, social media replies, and product descriptions. On one hand, AIGC can greatly reduce the time required to create marketing texts, promoting consistency across different content and ensuring a unified writing style and format for efficient content creation. On the other hand, AIGC can enhance the effective utilization of data by interpreting abstract data sources such as text, images, and different structures, helping marketing departments overcome challenges posed by unstructured, inconsistent, and unrelated data. Meanwhile, AIGC can achieve personalized product discovery and search through multi-modal inputs such as text, images, and voice, leveraging in-depth understanding of customer profiles to better meet consumer demands.
2. Stimulating the Flywheel Effect to Enhance Total Factor Productivity
In the scenario described by the flywheel effect: to make a stationary flywheel rotate, sufficient initial energy must be consumed to push the flywheel into motion. Once the flywheel reaches a certain rated speed, its inherent momentum and kinetic energy allow it to overcome significant resistance and maintain its motion. At this point, only a small amount of energy is required to drive the flywheel’s high-speed rotation. The flywheel effect of AIGC refers to its ability to leverage the three elements of “data + computing power + algorithms” to gradually achieve a virtuous cycle of continuously optimizing resource allocation and forming innovative technologies through the integrated capabilities of foundational large models, thereby promoting the enhancement of total factor productivity in the process of productivity improvement and technological advancement. According to the pathways of AIGC, the flywheel effect can be divided into “production flywheel” and “innovation flywheel”.
On one hand, AIGC can directly influence product production, intelligently operating product design, manufacturing, quality inspection, and finished product stages, driving the “production flywheel” and enhancing production efficiency to promote total factor productivity. For example, in the software engineering field where AIGC is most widely applied, AIGC can significantly reduce the time required for numerous tasks such as generating initial code drafts, code corrections and refactoring, root cause analysis, and generating new system designs. AIGC can accelerate the coding process, pushing the skill combinations and capabilities required for software engineering toward code and architecture design, significantly enhancing work efficiency. Data shows that software developers using Microsoft GitHub Copilot complete tasks 56% faster than those not using the tool. On the other hand, AIGC can drive the potential for technological automation, accelerating the accumulation and formation of knowledge, stimulating the “innovation flywheel”, paving the way for subsequent innovations, and thereby achieving an expansion at the frontier of production. In July 2021, DeepMind released the first artificial intelligence model, AlphaFold, that predicts protein structures from amino acid sequences. Within a year, under the operation of AIGC, this model achieved complete predictions of almost all known cataloged protein structures in science. Months later, the first new drug entirely discovered and designed by artificial intelligence entered phase II clinical trials. AIGC is rapidly driving the generation of innovative content, with the potential to achieve a leap in productivity in the future, raising total factor productivity to new heights.
(3) The Negative Impacts of AIGC on the Development of the Manufacturing Industry
While AIGC improves resource allocation efficiency, promotes total factor productivity, and releases economic potential, it also raises issues such as data bias and regulatory crises, which can negatively affect individual enterprises in the manufacturing industry and even the macro development of manufacturing.
-
Data Bias and Data Errors
The data bias and data errors in AIGC affect the effectiveness and accuracy of model predictions, hindering the protection and enhancement of individual enterprises’ rights and operational efficiency. On one hand, during the operation of AIGC, biases contained in the training data resources may be absorbed, echoing, automatically perpetuating social biases, prejudices, and discrimination, thereby excluding specific categories of groups in the model. On the other hand, the negative impacts caused by data errors are amplified; AIGC may increase the risk of unintentional dissemination of false information or malicious rumors. For instance, utilizing AIGC to combine text, images, videos, and even sounds from specific scenarios can inadvertently create erroneous information or intentional deception that causes substantial harm at the individual level, weakening societal trust in the information ecosystem at a broader scale, thereby affecting the operational mechanisms and efficiency of economic systems.
2. Structural Misalignment of Market Supervision
The application of AIGC in the manufacturing sector may lead to structural misalignment between development and supervision, exacerbating regulatory crises regarding the knowledge copyright, ownership, and other rights of AIGC-generated content, which is not conducive to the macro high-quality development of manufacturing. First, regarding the data copyright issues involved in AIGC training models, they often include a large amount of copyrighted data used without authorization from rights holders. Does unauthorized use of copyrighted data constitute infringement? If it does, should there be exceptions allowing the training of machine learning models? In May 2020, the World Intellectual Property Organization’s Secretariat explicitly discussed this issue but did not arrive at an authoritative consensus on regulatory handling. Additionally, whether new results generated by artificial intelligence can be copyrighted or patented, and if so, who can apply for it? Some argue that works autonomously generated by artificial intelligence do not enjoy copyright. However, the European Court of Justice (ECJ) stipulates that if a work reflects the “intellectual creation of the author”, copyright can be attributed to the person who laid the foundation for the machine’s creation. Overall, the regulation of AIGC still has many shortcomings and deficiencies, which to some extent affects the comprehensive and systematic trend of intelligent transformation in the manufacturing industry.
4. Constraints on the Transformation of Intelligent Manufacturing Driven by AIGC
The previous text defined the concept of AIGC and divided its industrial structure into foundational, operational, and application layers, clarifying the three main technical-economic characteristics of AIGC: self-adaptability, operability, and parallelism. Based on this, the author clarified the deep mechanisms of AIGC’s integration into the manufacturing industry, exploring the logical paths for macro high-quality development and the promotion of micro-operational efficiency in manufacturing, analyzing the negative impacts of AIGC on industrial development. The author will further explore the constraints on AIGC-driven intelligent transformation of manufacturing from the perspectives of foundational, operational, and application layers.
(1) Foundational Layer: Incomplete Infrastructure Construction Constraints
The construction of data integration and computing power infrastructure at the foundational layer is a prerequisite for AIGC to drive intelligent transformation in the manufacturing industry. Since May 2020, when the Ministry of Industry and Information Technology issued the “Guiding Opinions on the Development of Industrial Big Data”, clearly proposing the construction of a national industrial Internet big data center to support upstream and downstream enterprises in advantageous industries to open data, the integration and supply capabilities of data resources in China have significantly improved. According to IDC statistics, China’s data scale is expected to grow from 18.51ZB in 2021 to 56.16ZB in 2026, with a compound annual growth rate of 24.9%, ranking first globally. At the same time, the scale of computing power in China has achieved rapid growth; in 2021, China’s intelligent computing power reached 155.2 EFLOPS (exaflops), with corresponding supplementation of computing power infrastructure. However, the development of the AIGC industry, the diversification of AI scenarios, and the explosive growth of data have led to a significant increase in computing power consumption, posing new challenges to data integration and computing power fundamentals. Overall, the development of China’s AIGC industry still lacks complete foundational infrastructure, which is not conducive to the efficient operation of intelligent industries.
On one hand, inadequate data foundational construction has made the quality and quantity of data the primary factors constraining AIGC’s promotion of intelligent transformation in the manufacturing industry. First, with the rapid growth of data volume, the high-dimensional and diverse modal formats of data have become increasingly prominent, complicating AI modeling and involving multiple variable dimensions such as time and space, leading to exponential changes in computational complexity and increased difficulties in data labeling and management costs. Second, massive data inevitably brings greater data noise issues and risks of data bias, making the optimization of data quality challenging and posing greater challenges for models to effectively utilize data and realize the value of data elements. Currently, China’s foundational construction related to data integration and processing is still at an early stage, with significant room for improvement in controlling the “quantity” and ensuring the “quality” of data. First, the phenomenon of “emphasizing volume over quality” is prominent; while the scale of data center construction in China is rapidly expanding, the follow-up quality of construction is being neglected, leading to inefficiencies, high security risks, and high energy consumption in data storage and processing capabilities, resulting in low effective utilization rates of data centers. Second, the issue of “emphasizing construction over protection” is significant; the disaster recovery coverage and investment for data infrastructure in China are 34% and 2%, respectively, lower than levels in developed countries, indicating that the data protection capabilities of China’s data infrastructure need further enhancement, and the construction of data security disaster recovery systems still needs to be improved, falling short of the standards for “ensuring quality and quantity, maintaining safety” in data foundational construction, which is not conducive to the intelligent operation of the AIGC industry.
On the other hand, incomplete construction of computing power fundamentals and uneven development of computing power have become key factors hindering AIGC’s promotion of intelligent transformation in manufacturing at the foundational layer. In a narrow sense, computing power infrastructure refers to the infrastructure that provides computing power resources, primarily focusing on computing power resources, including foundational facilities, computing resources, management platforms, and application services, covering supercomputing centers, data centers, and intelligent computing centers that can provide diverse services. In a broad sense, computing power infrastructure refers to ICT services that integrate computing power production, transmission, and IT capabilities. In recent years, the rapid development of computing power in China has been notable, but the overall level of construction remains incomplete, and uneven development restricts further development of the AIGC industry. Specifically, first, the development of computing power infrastructure is uneven, with a tendency to “emphasize hardware over software” becoming apparent. Statistics show that large data centers in China are showing a rapid growth trend, accounting for over 80% of data centers. The layout of hardware construction is steadily advancing, with the number of government data open platforms increasing from 3 in 2012 to 193 in 2021, a 63-fold increase. However, software development is lacking, requiring the establishment of a complete development software stack to meet enterprises’ needs for bottom-layer compilation and integration scheduling and unified operational management for different training inference data formats and scales. Second, the distribution of computing power infrastructure development is uneven, with varying levels of computing power development across different regions. China’s computing power infrastructure mainly includes four forms: supercomputing centers, intelligent computing centers, data centers, and urban brains, which are currently under simultaneous construction. As of the end of 2022, 11 cities in China have supercomputing and intelligent computing centers, while most regions are still in planning and construction stages, indicating significant room for improvement in computing power infrastructure construction.
(2) Operational Layer: Non-specific Adaptation of Algorithm Models Constraints
The algorithm models and technology applications at the operational layer are the core driving forces for AIGC to promote intelligent transformation in the manufacturing industry. With the development of the digital economy and technological advancements, artificial intelligence algorithm models exhibit characteristics of diversification, massiveness, and specialization, with the foundational development of algorithms playing an important role in achieving intelligent transformation in manufacturing. Green efficiency and strong applicability have become primary demands. As of October 2023, the number of large models developed in China and the United States accounts for over 80% of the global total, with 79 large models having over 1 billion parameters in China, second only to the United States. Currently, how to adapt operational layer algorithm models to the actual application in the manufacturing industry has become a major issue in promoting intelligent transformation in manufacturing.
On one hand, general-purpose algorithm models are not adapted to actual application scenarios, as the diversification and complexity of application scenarios increase the operational difficulty of models. In the manufacturing industry, industrial tasks and scenarios are complex and variable, with significant differences in industrial mechanisms, production processes, and product technologies across different industries. Although AIGC can significantly enhance economic operational efficiency and total factor productivity, the current algorithm models of AIGC cannot meet the scenario adaptation needs at various stages of the manufacturing industry. For example, Siemens collaborates with an industry AI giant to empower product quality inspection through the deployment of Microsoft Azure Machine Learning and Siemens Industrial Edge solutions, aiming to expand quality through a simple and convenient approach. However, the highly fragmented industrial scenarios cannot be individually adapted to algorithm models within a short time; solving one scenario issue typically requires deep integration of multiple tasks, involving multi-task unified modeling and other issues, thus posing higher challenges for algorithms. Industrial digitalization is a trillion-level market, as well as a combination of trillion-level markets. For segmented fields, it is difficult to have sufficient available data to train large models from the pre-training stage, and general-purpose large models cannot adapt to the scenario demands of focused segmented fields, creating a natural contradiction that hinders the development of large models. Therefore, the operational layer needs to produce a large number of algorithms or models tailored to different scenarios and tasks, which not only exacerbates the workload and investment in repetitive work but also makes the issues of AI development thresholds and research and development efficiency increasingly prominent.
On the other hand, the development of algorithm models is not adapted to economic conversion capabilities; if technology cannot be effectively converted into economic output in a short time, cost pressures will increase. When algorithm models and technologies at the operational layer move from laboratory environments to manufacturing enterprises’ production environments, due to differences in the research environment and development objectives, enterprises focus more on the efficiency and low costs of technology investments, making economic conversion capabilities a decisive reference for evaluating algorithm models. However, the traditional repetitive development and training of unidirectional task models currently lead to increased costs, becoming a key obstacle to AIGC’s empowerment of intelligent transformation in manufacturing. China is a major manufacturing country; in 2021, China’s industrial added value reached $4.87 trillion, accounting for 29.8% of the global total, but the proportion of IT spending was only 14.64%, indicating a serious mismatch with the scale of the industry. Insufficient economic conversion capabilities limit enterprises’ research and application of emerging algorithm models, resulting in a low penetration rate of industrial intelligence in China. According to Capgemini statistics, the AI application penetration rate in top European manufacturing enterprises exceeds 30%, while the AI penetration rate in manufacturing enterprises in China is only 11%. Overall, there is still a certain gap between the intelligent development of China’s manufacturing industry and the international leading level, and the adaptation of algorithm models has become a challenge that must be overcome in the intelligent transformation of manufacturing.
(3) Application Layer: Insufficient Connection of Industrial Operations Constraints
The intelligent operation and connection capabilities of the application layer industry are important guarantees for AIGC to promote intelligent transformation in the manufacturing industry. For manufacturing enterprises, the intelligent transformation system is vast and lengthy, involving a wide range of areas, making the connection between different businesses or departments particularly important. Data from the intelligent manufacturing evaluation public service platform shows that as of 2022, only 4% of manufacturing enterprises in China have achieved connections and intelligent operations across various links, reaching a maturity level of four or higher for deep intelligence; 12% of manufacturing enterprises are progressing toward a level three maturity, in a state of networked integration and single-point intelligence; and 53% of manufacturing enterprises are still at levels one or two of intelligence maturity, lacking the capacity for systematic intelligent operations. In summary, insufficient connection capabilities of industrial operations are a key issue constraining intelligent transformation in manufacturing, primarily for two reasons.
First, the lack of a systematic AIGC management and application framework within the industry hinders the pace of intelligent transformation. First, there is a lack of data management mechanisms and guarantees. Due to the lack of overall planning in the foundational layer data integration process, many manufacturing enterprises have not established effective data management mechanisms and guarantees, leading to widespread issues such as complex data sources, uneven data quality, and inconsistent data standards, resulting in unidentifiable data, redundant data, insufficient data accuracy, and overdue data timeliness, which in turn affects the generated content of AIGC, leading to discrepancies between data results and actual situations, making it difficult to support upper-level applications. Second, the limited data integration capabilities of manufacturing enterprises are evident. The systems within manufacturing enterprises are dispersed, with financial, distribution, supply chain, and other links each having their own data systems, and most enterprises have not established effective data integration mechanisms. Personnel utilizing AIGC for data analysis and applications need to spend a significant amount of time integrating and cleaning data from various internal and external systems, which not only affects the progress of business completion but also fails to meet operational efficiency needs of enterprises. Lastly, the insufficient application capabilities of algorithm models are notable. Currently, there are few links within manufacturing enterprises that effectively apply AIGC, with a lack of application scenarios for algorithm models, single analysis dimensions, simple and rigid forms, and insufficient application depth. How to fully match general large models with manufacturing field scenarios has become a challenging issue for promoting deep intelligent transformation in manufacturing.
Second, the shortage of human resources for AIGC within enterprises and insufficient talent reserves limit the operations and connections of intelligent transformation in manufacturing. Talent is key to the integration of AIGC and the manufacturing industry, and it is the most important asset for enhancing the international competitiveness of intelligent manufacturing. As the pace of transformation and upgrading in the manufacturing industry accelerates, the scale of the industry and the demand for personnel are increasing. The “2022 Intelligent Manufacturing Talent Development Report” released by the China Electronic Information Industry Development Research Institute in cooperation with Zhaopin shows that the number of job postings in the intelligent manufacturing field increased by 53.8% year-on-year in 2022, with year-on-year increases exceeding 50% from 2020 to 2022. As the complexity of AI scenarios increases, traditional single-skill technical talents can no longer meet the development needs of manufacturing enterprises, necessitating more composite innovative talents with both business capabilities and AI technology application abilities. The Ministry of Human Resources and Social Security predicts that by 2025, the demand for talent in the intelligent manufacturing field will reach 9 million, with an estimated talent gap of 4.5 million. Therefore, to overcome the constraints of operational connections in the application layer, more comprehensive talent investment in the construction of intelligent transformation in the manufacturing industry is needed in the future.
AIGC’s Realistic Path for Integration into the Manufacturing Industry
At this stage, China’s manufacturing industry faces certain thresholds in absorbing and applying AIGC technologies, data elements, and digital infrastructure construction, necessitating further improvement in foundational infrastructure construction, specificity in algorithm model matching, and deepening of industrial operational connections to cultivate a favorable ecosystem for manufacturing industry transformation and accelerate the pace of intelligent manufacturing transformation. According to the industrial structure layers of AIGC, the specific paths for AIGC’s integration into the manufacturing industry can be explored in the foundational layer, operational layer, and application layer.
(1) Foundational Layer: Building the Foundation for Intelligent Manufacturing
AIGC, as a representative of the new generation of artificial intelligence technology, is a key force in disrupting and reconstructing traditional manufacturing production methods and models, with its immense development potential and industrial synergy effects being important supports for cultivating new momentum, driving new development, and expanding new spaces.
In manufacturing, the essence of intelligent manufacturing is to apply the latest industrial engineering and digital network technologies (mobile internet, edge computing, big data, artificial intelligence, Internet of Things, etc.) to re-examine existing processes and production organization methods, achieving operational innovations in supply, marketing, design, and manufacturing, and comprehensively promoting enterprises toward intelligent production, intelligent management, and intelligent operations to meet customers’ agility, personalization, and service needs. AIGC has become the foundational base for creating intelligent manufacturing and a key driving force for intelligent transformation in the manufacturing industry. The evolution of digitalization and intelligence in manufacturing can be divided into four stages: automated production lines and production equipment, equipment interconnection and data collection, data interoperability and direct application, and data intelligent decision-making and control execution. In this process, the foundational construction and promotion of each stage are not independent of each other, and the process of intelligent transformation in manufacturing may simultaneously span multiple stages. Under the influence of AIGC’s technical-economic characteristics, traditional manufacturing factories can achieve complete intelligent manufacturing transformation centered around intelligent decision-making once their digital and intelligent levels reach a certain extent.
Currently, the specific paths for AIGC’s integration into the manufacturing industry at the foundational level mainly include two: first, improving foundational data construction by enhancing computing power and data capabilities to stabilize the foundational transformation of intelligent manufacturing. Second, developing initial intelligent applications based on data by directly integrating the general capabilities of foundational large models such as question answering and code generation to enhance the operational efficiency of traditional manufacturing industries. For example, Salesforce, Microsoft, ABB, and Yonyou have integrated large models into CRM, ERP, and production management software, helping enhance the data analysis, document management, knowledge question answering, and other auxiliary capabilities of professional software.
(2) Operational Layer: Innovating Algorithm Scene Development Momentum
At the operational layer, the flywheel effect and innovation-driven strategies of AIGC enhance the innovativeness and adaptability of various algorithm models, effectively supporting the intelligent transformation of manufacturing by meeting specific demands in certain links that traditional industries find difficult to satisfy. Large models represented by ChatGPT and Llama have opened the door to general artificial intelligence, making AIGC an important focal point for global economic growth and providing new space for intelligent transformation in manufacturing. Foundational models based on architectures such as Transformers and U-Net have become the basis for generative artificial intelligence entering the manufacturing field, while vector databases and MaaS have become important digital infrastructures. Driven by algorithm models, the economic benefits of various industrial and manufacturing scenarios are fully tapped, providing innovative momentum for the transformation of intelligent manufacturing.
First, by enhancing the generality and adaptability of large models in manufacturing scenarios, AIGC can fully utilize its recognition levels and model transferability in common scenarios, achieving the matching of algorithm models in basic tasks and specific industry tasks. In other words, model parameters and structures are continuously enriched during the learning process, with deep learning platforms serving as the core carrier for intelligent transformation in manufacturing, accelerating the soft and hard adaptation of various chips such as GPU, CPU, ASIC, and FPGA, thereby improving model training speed in areas such as custom optimization, unified hardware interfaces, and automated compilation to meet the demands of more fragmented industrial manufacturing scenarios.
Second, focusing on specific fields within manufacturing through fine-tuning and external knowledge bases can achieve innovation in segmented scenarios, adding new functions and services. For example, Authentise has launched 3DGPT, a question-answering database for additive manufacturing technology, by fine-tuning 12,000 scientific papers, enabling users to obtain answers to specialized questions such as “How to reduce the likelihood of defects when using powder stainless steel?”, significantly improving the efficiency of professionals in that field in handling related issues.
(3) Application Layer: Catalyzing New Emerging Models for Industrial Operations
AIGC is an important sign of the transition from artificial intelligence 1.0 to 2.0; it not only represents a representative technology but also an ecosystem, embodying opportunities for the innovative evolution of traditional industries. With the continuous advancement of digitalization, AIGC will catalyze new operational models in the manufacturing industry through technology integration and data penetration, forming highly intelligent “Digital Factories” and igniting a new wave of industrial revolution in manufacturing. Digital factories utilize advanced technologies such as digital technology, big data, artificial intelligence, and the Internet of Things to connect production lines and production equipment, achieving an efficient, automated, intelligent, and adaptive advanced manufacturing model. Under the influence of AIGC, this manufacturing model can gradually realize real-time perception, storage, analysis, decision-making, and control of data in areas such as design, production, quality, logistics, and environmental protection, helping to reduce costs, increase efficiency, enhance product quality, improve customer satisfaction, and create core value.
Specifically, the paths for AIGC’s integration into manufacturing to catalyze digital factories at the application layer include two: first, establishing dedicated large model settings from a macro perspective, building specialized large models in the manufacturing field starting from pre-training to lay the foundation for the intelligent transformation of various processes in the industrial chain. For example, the manufacturing large model AInno-15B released by Innovation QiZhi can achieve automation based on AIGC in industrial production stages, including applications for task scheduling of generative industrial robots (ChatRobot), generative enterprise private domain data analysis (ChatBI), and generative enterprise private domain knowledge question answering (ChatDoc), among others. Second, the implantation of functional subsystems within enterprises, including Product Lifecycle Management (PLM), Enterprise Resource Planning (ERP), Manufacturing Operations Management (MOM), Warehouse Management System (WMS), and Distributed Control System (DCS), etc. These subsystems collectively form the core of digital factories, and after comprehensive integration and fusion, they can form an emerging intelligent manufacturing innovation platform, thereby promoting the intelligent transformation of manufacturing.
Long press the QR code to download the full text
Source Journal: Journal of Xinjiang Normal University (Philosophy and Social Sciences Edition)
First Published on:2024-01-16 17:18:20