Author|Li Lihui"Former Governor of Bank of China"
Article|"China Finance" media integration contribution
ChatGPT, an artificial intelligence chat program launched by OpenAI, Microsoft's artificial intelligence lab in the United States, on November 30, 2022, exceeded the 100 million monthly active users mark within 2 months, becoming the most popular product in the digital world, known as the singularity of artificial intelligence technology development. The full name of ChatGPT is Chat Generative Pre-trained Transformer, which can be understood as "a pre-trained human-machine dialogue conversion program with generating ability". ChatGPT has the learning ability, logic ability and language ability based on super computing power, can learn, master and use all kinds of knowledge in the big database, can learn, understand and use human language and expressions, can distinguish contexts for human-computer dialogue communication, and can complete tasks such as writing emails, papers, copywriting, video scripts, code and translation. On February 7, 2023, Microsoft integrated ChatGPT into the Bing search engine and Edge browser to solve the load overload problem; GPT-4 was released on March 14, establishing a text-to-image mapping relationship, supporting applications such as picture input, table input, programming with pictures, and writing based on images.
ChatGPT drives the expansion of AI applications worldwide. Google's Bard uses artificial intelligence to infuse generative AI into email, image editing and online work tools, previously only available to the US and UK, but now expands to 180 countries. On March 16, 2023, Baidu launched its generative AI product "Wenxin Yiyan", which can be used in scenarios such as commercial copywriting, literary writing, mathematical calculation, Chinese understanding, and multimodal generation.
The data used by OpenAI for ChatGPT learning and pre-training is essentially open and shared, with an overall uninterrupted data feed. The key factor that ChatGPT can stand out from its peers is that OpenAI has invested huge financial resources and manpower to cultivate super computing power since 2015, and finally formed the technical advantages of AIGC (AI Generated Content). Someone recently did a test, selected 4 types of classic test tasks to assess human rational thinking, including semantic illusion tasks, cognitive reflex tasks, falsification selection tasks, mental procedural tasks, a total of 26 questions, ChatGPT test, GPT-3.5 accuracy rate is 58%, GPT-4 accuracy rate is as high as 88%, higher than the average human accuracy rate of 62%.
The results of ChatGPT should prove that the innovation and application of artificial intelligence technology has entered a new stage of iterative upgrading. AI iterations can certainly create business value for developers, such as the current ChatGPT Plus user has to pay $20 a month, and OpenAI's market valuation has climbed significantly. As for the contribution of AI iteration to the digital economy and society, it needs to be evaluated and measured by economists.
What we must pay attention to is the possible impact and challenges of digital technology changes revealed by AI iteration on the development of the mainland's digital economy.
First, computing power concentration and computing power competition.
Computing power mainly refers to the ability to calculate data. It is generally believed that in the era of digital economy, data becomes a factor of production, and computing power constitutes a new productive force. The computing power as an infrastructure includes hardware with servers as the backbone and software with algorithm programs as the core. In 2022, the total computing power of the mainland will reach 180 Ea per second (1.8 trillion billion times/18,000 times) FLOPS (floating point arithmetic), which is said to rank second in the world after the United States.
In the application of digital technology, artificial intelligence, big data, and cloud computing all require huge computing power support, and artificial intelligence computing power reflects the most cutting-edge computing power. Hash power also largely determines financial competitiveness. In the financial field, intelligent credit evaluation, customer screening, risk pricing, risk control, investment advisory, actuarial insurance, digital employees, etc., all do not require the support of data and computing power.
The construction of computing power infrastructure requires continuous investment of huge financial and human resources, to afford to spend money, find the right people, and endure loneliness, which is bound to lead to the concentration of computing power. According to unofficial statistics, the computing power of the United States and China accounts for about 60% of the global computing power. The second is to concentrate on capital giants and technology giants, and the vast majority of small enterprises have insufficient financial and human resources to build valuable computing power.
Therefore, the competition for computing power will largely be national-level competition between major economies, as well as enterprise-level competition between capital giants and technology giants. What we need to explore is how to build a national computing power infrastructure that can occupy the global technological highland; How to build a computing power ecosystem with trusted technology, resource sharing, business sustainability, and market entities benefiting.
The second is the data divide and data governance.
Applications in different fields and scenarios have different performance and coverage, so they have different requirements for data elements. Not all applications require pre-training and require large-scale data. However, based on the competition of computing power at the national and enterprise levels, it will inevitably require data support at the national and enterprise levels. To build generative AI programs comparable to ChatGPT, you must need a disjointed, barrier-free data supply.
The problem is that the data gap derived from the administrative system, payment model, and geopolitics will pose substantial constraints on the mainland's computing power construction and computing power output.
Mainland China has the advantages of massive data scale and rich application scenarios, but the sharing mode of public data resources and mobile payment data is not perfect, which affects the in-depth development of the value of data elements. For example, credit data and behavioral data involving enterprises are scattered in different local systems such as financial regulatory departments, financial institutions, industrial and commercial administration, taxation, and customs, and the level of openness and sharing is not high, forming an administrative data gap. The scale of mobile payment users in mainland China is as high as 900 million, digital payment has become the main data entrance, and the Internet platform has ultra-large-scale personal data and enterprise data, but the data connection and data sharing between the Internet platform and financial institutions have not yet reached a mature model, and the data gap needs to be filled.
The United States has the world's leading economic strength and military strength, and scientific and technological strength is the basic support of US national strength. The United States occupies global data resources, including those of Western countries and most developing countries, which is the structural advantage of American scientific and technological innovation. In the geopolitical environment, the technical barriers set up by the United States and Western countries against China are constantly escalating, and it is likely to extend from high-end chips and core software to the field of data resources, artificially creating a data gap.
What we need to discuss is how to improve the data governance system, build a data infrastructure system that adapts to the characteristics of data, conforms to the development of the digital economy, and can ensure national data security, and fully realize the value of data elements; How to grasp the direction and principle of global data sharing, participate in cross-border data flow, and make full use of global data resources to create a competitive advantage in computing power while maintaining data sovereignty.
The third is AI synthesis and AI trust.
AI synthesis refers to the application of deep learning, virtual reality and other generative algorithms to produce deep synthesis content such as images, audio, video, and virtual scenes. As the level of AI synthesis simulation evolves, people are beginning to worry about the threat posed by AI fakes and AI manipulation to society.
According to statistics, on 10 mainstream platforms such as iQiyi, Tencent Video, Youku, Douyin, YouTube, and Twitter, 24,317 new deep synthetic video works were released in 2021, an increase of 13.5 times over 2017. The attention of deep synthetic content has grown exponentially, and the market momentum is abundant, and the newly released deep synthetic video has more than 300 million likes in 2021.
Deep synthesis content blurs the boundary between real and false, and the latest deep synthesis algorithms can resist general technical screening, and even make high-imitation immersive voiceprints. Rookie-level AI fakes have already been used to commit economic fraud, and may also be used to discredit individuals and businesses. Ash-level AI manipulation may be used to smear politicians or regime entities, manipulate negative public opinion, create political malice, undermine political trust, and intensify social conflicts.
Therefore, there is an urgent need to refactor AI trust. What needs to be explored is how to effectively combat AI falsehood in technology and system, how to establish a firewall to prevent AI manipulation, and maintain national security in the digital economy era.
In the face of the new situation of AI iteration and computing power competition, we must accelerate the construction of high-level computing power infrastructure and advanced data infrastructure systems, activate the potential of data elements, strengthen and optimize the digital economy, enhance new momentum for economic development, and build new advantages in national competition.
First, build a well-laid out world-leading computing power infrastructure.
In February 2023, the Central Committee of the Communist Party of China and the State Council issued the "Overall Layout Plan for the Construction of Digital China", which proposed to systematically optimize the layout of computing power infrastructure, promote the efficient complementarity and synergy of computing power between the east and west, and guide the reasonable echelon layout of general data centers, supercomputing centers, intelligent computing centers, and edge data centers. Improve the level of application infrastructure as a whole, and strengthen the digital and intelligent transformation of traditional infrastructure. Promote the aggregation and utilization of public data, and build a national data resource base in important fields such as public health, science and technology, and education.
The three major economic circles of the Yangtze River Delta, the Pearl River Delta and the Beijing-Tianjin-Hebei region are the gathering areas for data generation, data computing and digital technology talents, while some areas in the western and northern parts have the advantages of low electricity prices and low temperatures. The layout and construction of computing power infrastructure should anchor the world's leading goals, coordinate national and enterprise-level projects, integrate new centers and old centers, and take into account human resources and operating costs. I believe that the Yangtze River Delta Artificial Intelligence Supercomputing Center, Beidou Satellite Big Data Integration Application Industrial Base, Yangzhou Deheng Data Jiangsu Financial Industrial Park, and Tencent's cloud computing data center will all become benchmarks for computing power infrastructure.
Building computing power infrastructure requires specialized enterprises and professional teams. The organizers of the Yangzhou Forum, DeHeng Data, have the advantages of data center investment and construction, operation management, and R&D innovation, while Huawei and China Unicom have the advantages of information and communication infrastructure construction, software development, and mobile services. Such a number of scientific and technological innovation subjects, "soft and hard", "form and god", can play capable, love to fight and win, is China's pride.
Second, build a data element sharing system for high-quality supply and compliant circulation.
In early December 2022, the Central Committee of the Communist Party of China and the State Council issued the Opinions on Building a Data Infrastructure System to Better Play the Role of Data Elements, which formulated regulations for building a data infrastructure system and better playing the role of data elements, known as the "20 Data Articles".
The data basic system covers four aspects: data property rights system, data element circulation and transaction system, data element income distribution system, and data element governance system. Data property rights are defined as the right to hold data resources, the right to use data, and the right to operate data products, respectively establish mechanisms for granting confirmation rights to public data, enterprise data, and personal data, protect the legitimate rights and interests of all participants in data elements, improve compliance and supervision rules for the whole process of data, build standardized and efficient data trading venues, cultivate data element circulation and transaction service ecology, build a data security, compliance and orderly cross-border circulation mechanism, and complete the mechanism for data elements to be evaluated by the market and remuneration determined according to contribution. Give better play to the guiding and regulating role of the government in the distribution of data element revenues, and establish a secure, controllable, flexible and inclusive data element governance system coordinated by the government, enterprises, and society.
This is a guide for the construction of the basic system of the digital economy in the mainland, and we must seriously implement and implement it. First, follow the law of development, innovate institutional arrangements, and improve the market institutional mechanism of data elements; Second, adhere to sharing and sharing, release value dividends, and enhance the sharing and inclusiveness of data elements; The third is to strengthen high-quality supply, promote compliant circulation, and improve the quantity and quality of data element supply; The fourth is to improve the governance system, ensure safe development, and actively and effectively prevent and resolve various data risks; Fifth, deepen open cooperation, achieve mutual benefit and win-win results, and actively explore new ways and models for cross-border data flow and cooperation.
Third, build a credible, reliable and controllable digital security system.
Digital technology innovation drives economic development and social progress, and also brings risks and challenges to national economic and social security. For example, data fraud and information manipulation based on artificial intelligence technology, decentralized finance based on blockchain architecture, and malicious Trojans and technical dependencies based on open source software.
We must be highly vigilant against AI falsehood and AI manipulation. The focus is to improve the deep synthetic content identification technology, timely detect and prove AI falsehood, and provide public services against AI falsehood. Establish an AI trust system, explicitly prohibit AI fraud and AI fraud at the legislative and law enforcement levels, give qualified enterprises AI trust marks, establish firewalls at the national level to prevent AI manipulation, and maintain national security in the digital economy era.
Decentralized finance is not only a hot spot for global financial regulation in the future, but also a hot spot for international financial competition in the future. It is necessary to deeply analyze the existing and potential "subversive" performance of digital technologies such as distributed peer-to-peer architecture and decentralized architecture, focus on the possible paths for decentralized financial tools to cross financial infrastructure barriers and monetary and financial sovereignty boundaries, study technical countermeasures and policy plans, and build digital financial security barriers.
Software open source is penetrating more and more software products, and trusted open source can become a viable path for digital technology innovation. In the open source landscape, first, it is necessary to establish software security technical standards, establish a software audit and certification system, and prevent open source software with security risks and even malicious Trojans from entering the mainland; The second is to increase investment, increase policy support, encourage the mainland's own scientific research institutions, science and technology enterprises and digital technology talents to develop software with independent property rights, realize the optimization and balance of the basic performance, expansion performance and security performance of core software, and promote the progress and growth of the mainland software industry.