The following report by a Chinese state-run think tank describes how the Chinese government and foreign governments are using large AI models. The authors list more than 50 large models developed by Chinese tech companies that provincial and local governments in China have deployed for various purposes, and provide four brief case studies. While these AI systems have the potential to improve Chinese local governments’ provision of public services, the focus of many of these governments on early detection and suppression of social disturbances suggests that these AI models are being used to enhance and intensify China’s surveillance state.
An archived version of the Chinese source text is available online at: https://perma.cc/8SYR-Z2LN
Research Report on Governance Modernization in the Digital Age: Practice and Prospects for the Application of Large Models in the Government Domain (2023)
Preface
Currently, large generative artificial intelligence (AI) models have become a focus and hot topic for all different people. Large models such as ChatGPT and Bard are a leap forward in AI technology and represent the transition of AI technology from perceiving and understanding the world to creating the world. The government domain involves a large volume of content production and human-to-human interaction. It is highly aligned with the high-level information collection, text summary, and intelligent interaction capabilities of generative AI large models. This field is fertile ground for the future application of generative AI large models.
Since the emergence of large model tools, countries around the world have explored the possibility of applying new technologies in government governance, promoting the formation of a wave of digital government reforms with intelligentization (智能化) as their main feature. In terms of the scope of application, the application of large models in the government domain has been widely explored. Eighteen countries or regions, including the United States, the UK, Australia, Canada, Singapore, Japan, and South Korea, have applied large models in government affairs management, covering 13 specific scenarios in the 5 major fields of government internal office work, government information disclosure, government service provision, optimization of people’s livelihood (民生) services, and national defense and aerospace. In terms of application depth, the awareness and utilization rate of large models in the government domain have reached a high level in some countries. The Roland Berger consulting firm predicts that, provided they receive ample applications, large models are expected to reduce operating costs in the public service industry by 1.8%. In terms of deployment and promotion, countries and regions such as Singapore, Japan, and the United States are at the forefront in practice. They have promoted the transition from local and dispersed exploration to integrated applications and have made overall arrangements for deployment methods, data processing, and other aspects.
In order to make good use of the “double-edged sword” of large model technology, countries are actively promoting organizational, talent, and technological changes to adapt to new governance challenges, while accelerating the application of large models in the government domain at the same time. Since 2023, South Korea, Singapore, the United States, the UK, Australia, New Zealand, Japan, Canada, and Denmark have successively issued interim guidelines for the use of generative AI by government agencies (civil servants). They are working to avoid technical risks and promote the compliance of applications by clarifying usage principles, framing the scopes of scenarios, and establishing regulatory norms. However, because the technology is still in the stage of rapid development, ambiguity still exists in governance policies: First, a complete risk governance framework for government application of AI technology has not yet been formed. Second, scenario categorization and grading, and end-to-end supervision specifications are relatively rough and cannot achieve full coverage of new technology application behaviors.
In China, the application of large models in the government domain is still in its initial exploratory stage. According to statistics, there are at least 56 large model vendors in China that have deployed products in the government domain. Of these, 15 vendors including Baidu (百度), Beijing Knowledge Atlas Technology Co., Ltd. (Zhipu AI; 智谱华章), iFlytek (科大讯飞), SenseTime (商汤科技), the Institute of Automation of the Chinese Academy of Sciences (CAS Institute of Automation; 中国科学院自动化研究所), Alibaba Cloud (阿里云), 360, and Kunlun Tech (昆仑万维) have registered large model products. Localities including Beijing, Shanghai, Hangzhou, and Shenzhen have introduced policies to promote the implementation of large model applications in the government domain. Government hotlines, intelligent customer service, urban governance, healthcare, and education are key implementation directions. With the rapid development of large generative AI models, it is expected that, in the future, related technologies will be widely applied in the construction of smart cities and the provision of government services in various places, becoming a powerful force promoting a new round of governance transformation. Government agencies need to strengthen forward-looking deployment, proactively respond to the opportunities and challenges brought by new technological changes, develop a full understanding of large model technology from multiple levels, including strategy, business, organization, and risk, accelerate explorations into the feasible paths for the development and application of large government models suitable for China, and accelerate the modernization of government governance.
To view the rest of this translation, download the pdf below.