Since the beginning of this year, government units in many places such as Shenzhen, Hangzhou, and Suzhou have announced access to the DeepSeek (In-depth Search) big model to promote the application of AI in the field of government affairs, and "AI+Government Affairs" has attracted attention.

  "AI is gradually embedded in scenarios such as government office, urban governance, people's livelihood services and auxiliary decision-making." Tian Feng, director of Quicksense and Slowsense Research Institute, said that "AI+Government Affairs" can improve decision-making efficiency. For example, Hubei Lichuan has achieved functions such as second-level retrieval of policy documents and automatic correction of official document formats through the deep integration of the DeepSeek big model and the OA (office automation) system, greatly improving the efficiency of government documents.

  "AI+Government Affairs" can also help precise environmental control. With the help of multi-modal large models, combined with satellite remote sensing, drone monitoring and public vehicle cameras, the perception and analysis of urban space can be realized, and used for environmental protection and sanitation, garbage supervision, land conflict handling, etc.

  However, there are also some problems in the field of "AI+Government Affairs". Yang Haiming, chief technology officer of China Unicom Digital Technology Co., Ltd., admitted that at present, the problem of "using AI for the sake of using AI" is more serious, and the phenomenon of "using AI for the sake of using AI" is more common. For example, some departments announced that they had completed the intelligent work after only accessing the full-blooded DeepSeek big model, but did not seriously consider the subsequent improvement of the big model's capabilities and in-depth application scenarios. Tian Feng also bluntly stated that due to the lack of long-term strategic planning, the phenomenon of "sports" promoting AI also exists. Some departments lack the ability to continue operating "AI+Government" products.

  The data collected by some government departments are not of high quality and are not updated in a timely manner. Some of the collected data are mostly centered on themselves, and do not start from market demand, nor do they explore the value of data based on market demand, resulting in a lot of data being idle and wasted. Tian Feng gave an example and said that traffic accident site data is very valuable to smart driving companies, but this type of data is usually only in the hands of government departments and is less fed back to related companies. At the same time, this data will be cleaned regularly, resulting in the failure of data assets to achieve high value.

  Tian Feng suggested that government units formulate long-term strategies covering the application of "AI+Government Affairs", clarify specific action plans, and ensure effective implementation of policies. At the same time, positions similar to chief AI officers will be established to give full play to the role of young innovative talents and improve the AI ​​literacy of civil servants. Yang Haiming suggested adhering to problem-oriented and demand-oriented approaches. On the one hand, focus on practical problems and form an effective closed loop for the application of "AI + government affairs"; on the other hand, focus on actual needs, conduct in-depth research on government-side efficiency and people-side needs, and find the next development direction.

  Regarding the data problem, Yang Haiming suggested that the country or open source community should build some standard data sets as the basis to check ideology from a technical level, "We should provide AI with standardized high-quality data just like providing students with standardized textbooks."

  When talking about data infrastructure construction, Tian Feng suggested integrating infrastructure resources such as computing power, network and energy, and building AI computing power centers and data platforms based on cities or multiple cities to realize the sharing and efficient utilization of data infrastructure resources.

  "When promoting the application of AI in government affairs, we can first pilot high-frequency and low-risk systems, then iterate to medium-low-frequency and high-risk systems, establish a horse racing mechanism, and promote experience sharing and replication." Tian Feng emphasized, "We must encourage innovation and tolerance of trial and error."

  Yang Haiming emphasized that from privacy and security to ethical risks, from sensitive data leakage to large-scale model hallucinations leading to wrong judgments, these potential risks are also issues that must be considered when AI is used in government affairs.

[Editor in charge: Zhou Jingjie]