After WeChat’s revision, how to quickly find us
1. Click the blue text “Enterprise Development and Reform” below the title
2. Click the “…” in the top right corner of the page
3. Click “Set as Starred”
Abstract
The AI large model is accelerating the arrival of the third wave of “digital-physical integration”. As a new productivity tool, it will inevitably expand from the content field into the physical production field, triggering a new efficiency revolution in various aspects of manufacturing, and accelerating the intelligent transformation of the manufacturing industry.
Currently, we must seize the historical opportunity presented by the new generation of artificial intelligence technologies represented by AI large models, utilizing the comprehensive empowerment and support of the technology system represented by “public cloud + AI” to accelerate the advancement of intelligent manufacturing to a new stage.
The AI large model is a significant milestone in the development of general artificial intelligence, characterized by General Purpose Technology (GPT), and is accelerating the third wave of “digital-physical integration”. Large models have restructured the fundamental ways humans retrieve, create, and apply knowledge, upgrading to support and empower knowledge workers. When integrated with product research and development, process design, production operations, and various specific scenarios in manufacturing, they can form new productivity, thereby deeply influencing the research and innovation modes in manufacturing, and accelerating the intelligent transformation and upgrading of the manufacturing industry.
1. Digital-Physical Integration is an Important Watershed in the Global Manufacturing Competitive Landscape
Digitization is a watershed in times of great change and has become a catalyst for the sharp differentiation of competition among enterprises, cities, and countries. Manufacturing is an important sector for digital-physical integration, and the manner, breadth, and depth of this integration can directly influence or even determine the advanced level of manufacturing and the global competitive landscape.
(1) Digital-Physical Integration is the Fundamental Reason for the Global Leadership of American Manufacturing
Currently, many subconsciously endorse the development path and model of German manufacturing in the digital age, while “singing the decline” of American manufacturing. However, in the past decade, the U.S. has been a “model room” for global manufacturing development. Whether in terms of the scale, growth rate, or competitiveness of the manufacturing industry, the U.S. has consistently led Germany and Japan, with the gap constantly widening: from 2011 to 2021, the scale of U.S. manufacturing expanded from 1.5 times that of Germany to 2.4 times, and from 2.5 times that of Japan to 3.1 times, with U.S. manufacturing still growing faster than both Japan and Germany. The consolidation and establishment of the U.S. manufacturing leadership position over the past decade is the result of “software defining hardware”, the deep and comprehensive integration of digital technologies represented by “cloud + AI” into the real economy, and the migration to new digital infrastructure.
(2) The Emergence of Digital Native Enterprises is an Important Sign of Upgrading American Manufacturing
The emergence of digital native enterprises is an important sign of industrial upgrading and economic prosperity. The widening gap between German, Japanese, and American manufacturing is primarily manifested at the micro level by the lack of a competitive batch of digital native enterprises in the manufacturing sector. Germany’s “Industry 4.0” goal has not been achieved, and the gap from expectations is significant, with the digitalization progress of small and medium-sized enterprises being slow; research shows that only 21% of small and medium-sized enterprises have used digital technologies in production, and a competitive batch of digital native enterprises has not yet emerged. Japan’s situation is similar to Germany’s; Japan has experienced a “lost two decades” and also failed to cultivate a group of digital native enterprises.
In contrast, the U.S. sees a continuous emergence of new enterprises and products, with digital native companies such as Tesla, SpaceX, Rivian, OpenAI, Snowflake, and Palantir not only becoming global industry leaders but also continuously constructing new models of product innovation. For example, the U.S. B-21 bomber adopted digital design from the outset, utilizing cloud computing for full lifecycle R&D, becoming the fastest developing model in the U.S. military in the last 30 years, and can continuously upgrade product functions like Tesla cars by downloading new software. The essence behind this is that “cloud + AI” has become not just a commercial infrastructure but also an innovation infrastructure, a cradle for nurturing and incubating new enterprises and products.
(3) AI Large Models are a New Starting Point for Reshaping the Global Manufacturing Competitive Landscape
AI large models are accelerating the comprehensive arrival of the third wave of “digital-physical integration”, with intelligence as its main feature. AI large models will influence the development pattern of the manufacturing industry, primarily because they will affect the R&D and innovation modes in manufacturing. From current industrial practices, AI large models can not only directly empower the R&D innovation of intelligent vehicles, robots, chips, clothing, etc., such as engineers automatically generating code instructions through large models to complete the development and debugging of robot functions, but can even create entirely new functions for robots. More importantly, AI large models may also empower foundational research in manufacturing, revolutionizing traditional research paradigms. For instance, in 2022, the AlphaFold2 model developed by DeepMind predicted almost all protein structures. Today, AI models can not only “predict” but also “generate” proteins, creating new possibilities for future drug production and development. For example, ProGen, developed by the U.S. company Scale AI, successfully generated entirely new proteins from scratch.
Looking back at the history of global manufacturing development, every change in the relationship between humans and objects signifies another qualitative leap in manufacturing levels and brings new transformations in the competitive landscape of manufacturing. The U.S. has achieved breakthroughs in AI ahead of this round of competition and has been the first to implement it in the industry. The core insights behind this are: (1) The systematic capability of “public cloud + AI” is the ticket for technological innovation and industrialization of AI large models. The online service model of public cloud can provide concentrated, online computing power infrastructure, which is a necessary condition for the large-scale industrialization of AI large models; (2) Building an open-source ecosystem for AI is the main battlefield for global competition of large models. The U.S. has a globally influential open-source ecosystem, with the number of open-source frameworks being more than nine times that of China. Currently, building integrated open-source ecosystem capabilities for model training, tuning, deployment, and management is an important foundation for achieving breakthroughs in foundational technologies of large models and constructing application ecosystems.
2. Four Basic Trends of AI Large Models Empowering the Manufacturing Industry
In the era of “software defining everything”, AI large models, as new productivity tools, will inevitably expand from the content field (text-to-text, text-to-image, etc.) into the physical production field, triggering a new efficiency revolution in various aspects of manufacturing and accelerating the intelligent transformation of the manufacturing industry.
(1) AI-Driven Software Upgrades are the Main Way Large Models Empower Manufacturing
Industrial software is a key support for intelligent manufacturing. In the AI era, whether it is R&D, management, production, or after-sales service industrial software, it is necessary to upgrade it using large models.
At the industrial software development level, AI large models are revolutionizing the software development paradigm. AI will collaborate with humans in development, exponentially increasing the efficiency of software R&D, such as content generation tools (documentation, coding, testing, publishing, operations) serving front-line R&D personnel, which can significantly enhance productivity. Additionally, the research and application of “code large models” are triggering a revolution in AI coding. Utilizing the knowledge reserves and generation capabilities of large language models (LLMs), the new generation of AI coding platform products built on code large models possesses powerful code understanding and generation capabilities, supporting core scenarios such as code completion, unit test generation, code explanation, and code error checking. According to estimates, the coding capability of GPT-4 is equivalent to that of a Google L3 engineer with an annual salary of $180,000. Furthermore, with the rise of MaaS (Model as a Service), the model-centric development paradigm will lower the threshold for industrial software development and improve development efficiency.
At the performance level of industrial software, AI large models will promote intelligent upgrades of software. For instance, in R&D design scenarios, Back2CAD has launched CADGPT™ based on ChatGPT, supporting intelligent recommendations, document generation, code production, and other functions, effectively assisting product R&D design. Currently, Alibaba Cloud is also trying to leverage the capabilities of AI large models to drive the intelligence of industrial software SCADA. SCADA systems (Supervisory Control and Data Acquisition) can be applied in various fields such as electricity, metallurgy, petroleum, chemicals, gas, and railways for data acquisition, monitoring control, and process control. In the SCADA scenario, Alibaba Cloud’s typical approach is to use large models to generate industrial logic code in specific industry scenarios through programming interfaces and ecological libraries, automatically integrating it into industrial software, and optimizing models based on result feedback loops.
(2) Bridging Data Flow Breakpoints is an Important Value of AI Large Models Empowering Manufacturing
Every breakthrough in human-machine interaction technology brings about an industrial reconstruction. AI large models have brought a new “human-machine interaction” revolution, where natural language will be able to control everything in the future, profoundly changing how people use search engines, shop, and manufacture, and deeply influencing the future competitive landscape of industries.
The natural language interaction capabilities based on AI large models provide new possibilities for real-time, ubiquitous connections within manufacturing enterprises and between upstream and downstream of the industry, helping to resolve the contradictions between the global optimization needs of enterprises and fragmented supply, thereby achieving end-to-end global optimization. For example, Guangdong’s Tuosida Robotics Company has developed a model for the robotics industry using the Tongyi large model, allowing someone with a basic understanding of code to automatically generate robot code through dialogue to complete tasks in different scenarios, not only avoiding data breakpoints, reducing the impact of manual intervention, and improving product stability and reliability, but also lowering technical thresholds and development costs, helping to achieve the widespread adoption and popularization of AI technology.
As we enter the digital age, the previously highly integrated and centralized manufacturing system is gradually moving towards decentralized production and flexible organization. In this context, collaborative efficiency is crucial. The combination of AI large models and intelligent collaborative office platforms helps to break through various data flow breakpoints in manufacturing, promoting the efficient flow of data in R&D, production, distribution, and service processes, thereby enhancing collaborative efficiency within manufacturing enterprises and even between upstream and downstream of the industry, driving the manufacturing industry towards “intelligent collaborative production”.
(3) Entering Control Links is a Key Mark of AI Large Models Empowering Manufacturing
The universality and generalization of AI large models, along with the new development paradigm based on “pre-training + fine-tuning”, will empower manufacturing across various links, including R&D design, production processes, operational quality control, sales service, and organizational collaboration. Among these, we believe that entering the core control systems in production, such as PLC, MES, SCADA, etc., and enhancing the intelligence of process production flows are key marks of AI large model applications in manufacturing. Siemens and Microsoft announced a collaboration in April this year to drive the next generation of automation technology transformation based on GPT, developing code generation tools for PLCs and integrating AI large models into control links. Currently, in the field of power dispatch, AI large models can deeply engage in the core business processes of complex dispatch control in new power systems, serving as “expert assistants” for dispatch operations, providing power dispatchers with grid control strategies, optimizing line load balancing, and reducing grid losses.
Taking the automotive industry as an example, the transformation of the automotive industry over the past few decades has not only been a power revolution but also a control revolution. The biggest technological change faced in the evolution from traditional to intelligent vehicles lies in the innovation of automobile control systems, shifting from over 80 electronic control units in traditional cars to a centralized architecture similar to smartphones (underlying operating system + chip SOC + application software). Today, autonomous driving has become another major direction of transformation in the automotive industry. Currently, the changes brought by large models to autonomous driving mainly have two directions: first, large models serve as empowering tools, assisting in the training and optimization of autonomous driving algorithms; second, large models enter the decision control link, acting as “controllers” to directly drive vehicles and change outcomes. Reports in August 2023 revealed that Tesla’s “end-to-end” AI autonomous driving system FSD Beta V12 was publicly unveiled, relying entirely on onboard cameras and neural networks to recognize roads and traffic situations and make corresponding decisions. Of course, the actual application and implementation process of AI large models entering control links still face many issues that require further exploration and resolution by researchers.
(4) Collaboration Between Large and Small Models is an Important Trend of AI Large Models Empowering Manufacturing
AI large models themselves are not products and cannot directly provide services; they need to find specific scenarios to truly land. Although general large models possess strong generalization, they are still distant from solving practical scenario problems across various industries. From the perspective of actual industrial development, an important trend is the collaboration of various models such as general/specialized and open-source/closed-source, which is a necessary stage for industrial implementation. Moreover, in this stage, the highly collaborative important carrier of large and small models—AI agents—will become new production tools.
AI agents generally refer to intelligent entities based on LLMs that can autonomously complete specific tasks using tools. AI agents will collaborate LLMs with other models, software, and external tools to handle various complex tasks in the real world. In the future, AI agents will mainly consist of a “perception system + control system + execution system”, possessing not only generation capabilities but also task understanding, task decomposition, task scheduling, execution planning, and chain collaboration capabilities. Among these, LLMs will primarily play the role of a command center, similar to the human “brain”, managing the unified intelligent scheduling of digital tools (such as SaaS software, industrial robots, digital humans, etc.) integrated with AI agents, collaboratively solving practical problems in specific scenarios in real-time during production, management, and service.
3. Build the Systematic Capability of “Public Cloud + AI” to Promote Intelligent Manufacturing to a “New Stage”
On May 5, 2023, the first meeting of the 20th Central Financial and Economic Committee emphasized the need to grasp the wave of new technological revolutions such as artificial intelligence, adapt to the requirements for harmonious coexistence between humans and nature, maintain and enhance the advantages of a complete and well-supported industrial system, efficiently gather global innovative elements, promote industrial intelligence, greening, and integration, and build a modern industrial system with integrity, advancement, and safety. Manufacturing is the foundation of the national economy; today’s transformation and upgrading of manufacturing are no longer a single technology’s empowerment but require comprehensive empowerment and support from the technology system represented by “public cloud + AI”. Currently, we must seize the historical opportunity presented by the new generation of artificial intelligence technologies represented by AI large models and accelerate the promotion of intelligent manufacturing to a new stage.
(1) Implement a “Public Cloud First” Strategy, Making Public Cloud a Key Force for Promoting the Integration and Innovation of “Manufacturing + AI Large Models”
The large-scale, highly available, and low-cost computing power infrastructure of public cloud has become a key foundation for industrial intelligence. Especially after the U.S. upgraded chip controls, the public cloud is the optimal path to alleviate the bottleneck of high-end chips, breaking through the single performance chip bottleneck by efficiently connecting heterogeneous computing resources to collaboratively complete large-scale intelligent computing tasks, effectively reducing dependence on overseas high-end chips. First, the “public cloud first” strategy should be regarded as an important part of relevant policy planning in the digital transformation of manufacturing, clarifying medium- and long-term development goals, key tasks, and guarantee measures; second, it is necessary to be wary of the fragmented construction of computing power centers by various regions, which could lead to a fragmented unified computing power market, avoiding the phenomenon of building many but not using well or being unaffordable; third, data center utilization efficiency and the proportion of public cloud should be used as assessment indicators for data center construction, reversing the current situation of “heavy construction, light operation” and “heavy investment, light performance” in data center construction.
(2) Encourage Open Source and Openness of Models, Supporting Technology Platform Enterprises to Grow Strong and Prosperous Open Source Communities, and Enriching the AI Industry Technology Ecosystem
The competition in AI is not only a technological war but also a commercial war, and its core is an ecological battle, with the key being open source and openness. Open source and openness can lower R&D costs and application thresholds, serving as a “booster” for innovation to commercial closed loops. First, the top-level design of the AI open-source and open ecosystem should be well done, incorporating the construction of the AI open-source and open ecosystem into national planning and ensuring its implementation; second, local governments should be encouraged to collaborate with leading AI open-source community platforms to build AI empowerment centers, accelerating the intelligent innovation applications of manufacturing relying on massive open-source models and model-as-a-service platforms (MaaS platforms); third, application traction should be encouraged to accelerate industrial implementation, supporting manufacturing enterprises in rapidly applying foundational large models, developing industry models, and enterprise-specific models, using “models” to feed back into technological innovation.
(3) Launch AI-Driven Upgrade Projects for Industrial Software, Accelerating the Intelligent Upgrade of Manufacturing Across All Links and Chains
As a key support for intelligent manufacturing, industrial software plays an important strategic role in promoting the transformation and upgrading of the manufacturing industry. In the AI era, all industrial software deserves to be upgraded using large models. First, we should vigorously develop AI-based industrial software, promote the R&D of “industrial software + AI large models” technologies, enhance the independent innovation capabilities of industrial software in the intelligent era, and actively promote the formulation of industrial software standards; second, we should fully leverage the communication bridge role of industrial software-related alliances, utilizing the advantages of AI enterprises, industrial software enterprises, research institutes, and manufacturing enterprises to build a cooperative and win-win AI-driven industrial software ecosystem with core competitiveness.
(4) Focus on Key Industrial Chains in Manufacturing, Creating Benchmarks by Segment and Scene, and Demonstrating the Scale Application of Large Models in Manufacturing
The key industrial chains in manufacturing are important supports for accelerating the construction of a modern industrial system. It is essential to identify key links, concentrate high-quality resources, and build an element system centered on “computing power + algorithms + data” to enhance the degree of digital-physical integration in manufacturing, promoting the safety and intelligent upgrading of the manufacturing industrial chain. First, initiate demonstration projects for new industrialization supported by large models, using AI large models as leverage to promote comprehensive and in-depth empowerment of new industrialization and accelerate the exploration of new models for new industrialization; second, in manufacturing industrial belts with good industrial foundations and strong innovation capabilities, as well as advantageous development zones and industrial parks, take the lead in carrying out demonstration projects for the integrated innovative development of “manufacturing + AI large models”; third, through models such as “innovation platforms + digital factories”, strengthen joint innovation in key links such as perception, control, decision-making, and execution, targeting weaknesses, creating innovation application benchmarks, and promoting the large-scale application of large models.
(The author is the director of the Alibaba Cloud Intelligent Technology Research Center. The original title of the article is: Manufacturing is the Main Battlefield for the Application of AI Large Models)
(This article was published in the 12th issue of the magazine “Enterprise Reform and Development”. For reprints, please indicate the source. The opinions in the article only represent the author’s views and are for readers’ consideration.)
Welcome to subscribe
