In order to promote the deployment and application of artificial intelligence big models in the field of government affairs in a safe, steady and orderly manner, the Cyberspace Administration of China and the National Development and Reform Commission recently jointly issued the "Guidelines for the Deployment and Application of Artificial Intelligence Big Models in the Field of Government Affairs", providing guidance and basic reference for the deployment and application of artificial intelligence big models for government departments at all levels.
Full text attached
Guidelines for the Deployment and Application of Artificial Intelligence Large Models in the Field of Government Affairs
1、 Overall requirements
Adhere to the guidance of Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era, thoroughly implement the spirit of the 20th National Congress of the Communist Party of China and the 2nd and 3rd Plenary Sessions of the 20th Central Committee, fully implement General Secretary Xi Jinping's important thought on building a strong cyber country, fully and accurately implement the new development concept, coordinate high-quality development and high-level security, adhere to systematic planning and intensive development, people-oriented, standardized application, joint construction and sharing, efficient collaboration, safety and stability, and strive for practical results, orderly promote the deployment, application, and continuous optimization of artificial intelligence big model technology, products, and services in the field of government affairs, fully leverage the advantages of artificial intelligence big models in complex semantic understanding and reasoning, multi-modal content generation, knowledge integration and analysis, etc., provide efficient assistance for staff, and provide convenient services for the public and enterprises, Promote innovative development of government affairs, enhance governance efficiency, optimize service management, and assist in scientific decision-making.
2、 Application scenarios
Government departments can explore and apply artificial intelligence models in typical scenarios based on common and high-frequency needs in government services, social governance, office work, and auxiliary decision-making, tailored to local conditions and practical situations. Mainly includes the following reference scenarios:
（1） Government service category
1. Intelligent Q&A. Integrate local, departmental, and domain business resources and knowledge base data, utilize natural language understanding, retrieval enhancement generation, and knowledge graph technologies to provide convenient online government consulting services, enhance accurate understanding of public demands, generate real-time reference answers, help solve public doubts, and improve the convenience of information acquisition.
2. Assist in processing. Integrating government service guidelines, common problems, user evaluations, and historical records, utilizing intelligent matching and automated processing technologies, we provide one-stop government assistance services such as intelligent guidance, personalized guidance, form pre filling, auxiliary review, progress inquiry, and reminders. We assist staff in efficiently reviewing materials and support the public and enterprises in handling matters conveniently.
3. Policy services are available for immediate enjoyment. Build a policy service knowledge base, refine policy requirements, policy labels, push conditions, application process and other related content, use the "policy finding people" and "policy finding enterprises" algorithm models, strengthen public and enterprise demand analysis, achieve intelligent policy matching, promote proactive and accurate delivery of services that benefit the people and enterprises, and provide one-stop processing.
（2） Social governance category
4. Intelligent monitoring and inspection. By utilizing devices such as drones, video surveillance, intelligent sensors, and computer vision technology, real-time analysis of monitoring videos, images, and IoT perception data is carried out to assist staff in monitoring infrastructure such as buildings, roads, gas, bridges, water supply, drainage, heating, and comprehensive pipe galleries in real time. Abnormal behavior, environmental problems, or facility failures are detected in a timely manner, potential risks and hidden dangers are automatically identified, timely reminders are given, and disposal suggestions are provided based on the severity and severity of the abnormal situation to improve monitoring and inspection efficiency.
5. Assist in law enforcement and supervision. By using technologies such as speech recognition, video analysis, knowledge graph, and logical reasoning, law enforcement personnel are assisted in real-time inputting case information into the system, penetrating problem clues, generating case reports, quickly retrieving legal basis and judicial interpretations, and querying similar typical cases. Targeted case handling suggestions are provided to improve law enforcement and regulatory efficiency and standardization.
6. Market risk prediction. By using generative time series analysis models and anomaly detection algorithms, various market data can be monitored and analyzed in depth to capture market trends, including fluctuations in economic indicators, abnormal situations, etc., predict possible market risks, assess the impact on the economy and society, and issue timely warnings, providing support for government management and social governance.
（3） Office related
7. Assist in drafting documents. By utilizing the generation capability of the big language model and building a local knowledge base and preset templates, we provide writing suggestions, assist in drafting documents, check, proofread, and optimize formats and content for staff, improve work efficiency, and reduce the burden on grassroots.
8. Data retrieval. By utilizing knowledge graph construction and information retrieval technologies, we can accurately understand the data retrieval needs of staff, achieve rapid retrieval, precise positioning, multi-dimensional sorting, intelligent correlation, and comparative analysis of government information, and help staff improve the efficiency and accuracy of data retrieval.
9. Intelligent distribution. By utilizing technologies such as natural language understanding and multimodal recognition, multi-dimensional task classification and assignment rules are constructed to automatically identify, accurately classify, assist in filling out, and prioritize tasks such as incoming documents, incoming calls, and work orders, achieving assisted distribution and intelligent assignment, and improving task assignment efficiency.
（4） Auxiliary decision-making category
10. Disaster warning. Perform big data correlation and comprehensive analysis and judgment on multi-source, multi-dimensional, and multimodal data from satellites, ground sensors, geological monitoring stations, as well as forecasting and warning, disaster risk surveys, etc., identify abnormal fluctuations, predict possible natural disasters, and issue early warnings to assist government departments in taking effective measures in a timely manner, reducing disaster risks, and minimizing disaster losses.
11. Emergency response. Using reinforcement learning and other technologies, analyze and judge the nature, characteristics, degree of harm, scope of impact, development trend, and public response of social public safety and other emergencies, timely discover and warn of risks and hidden dangers, quickly simulate the effectiveness of emergency response plans based on emergency scenarios, force resource distribution, etc., provide scientific and reasonable emergency response suggestions, optimize rescue resource allocation, and improve emergency response speed and efficiency.
12. Policy evaluation. Utilizing artificial intelligence models to infer and analyze capabilities and data mining skills, analyzing public feedback, market reactions, economic indicators, and social satisfaction, constructing multidimensional indicators, evaluating the degree of policy goal achievement, policy influence, and potential problems, and supporting policy formulation departments to optimize policies.
13. Intelligent assisted evaluation. By utilizing self-learning generalization, humanoid evaluation reasoning, multimodal intelligent analysis and other abilities, project evaluation is carried out in accordance with relevant requirements. The content of project documents is deeply scanned and intelligently analyzed, and evaluation opinions and suggestions are proposed to assist in improving project evaluation efficiency and scientificity.
3、 Standardized deployment
Government departments should fully demonstrate the application requirements, implementation paths, and functional design of artificial intelligence models based on their actual work and scenario characteristics, select appropriate deployment modes, coordinate implementation, promote co construction and sharing, and enhance construction management efficiency.
（1） Reasonably choose the implementation path
Government departments should carefully choose the implementation path of artificial intelligence big models based on the needs of different government scenarios and existing technological foundations. For scenarios with strong universality and abundant data resources such as intelligent question answering and auxiliary document drafting, mature model products and services on the market that have been registered with the cyberspace administration department should be adopted. For scenarios with strong professionalism and complex business logic such as auxiliary law enforcement supervision and market risk prediction, targeted training can be conducted using domain expert knowledge and professional data to create vertical models. On the premise of ensuring security and not disclosing state secrets, work secrets and sensitive information, make full use of Internet computing power and model resources to carry out deployment and application of artificial intelligence large models in the field of government affairs. Encourage the exploration of innovative applications such as government intelligent agents and embodied intelligence.
（2） Coordinated and intensive deployment
Government departments should carry out the deployment of artificial intelligence models in the field of government affairs in a coordinated and intensive manner, relying on the "East West Computing" and the national integrated computing power network, to promote the layout of intelligent computing infrastructure, and implement centralized and unified security management and systematic technical protection measures to avoid "fragmented" security risks. Central and national government departments, as well as provinces (autonomous regions, municipalities directly under the central government) with conditions, can deploy intelligent computing resources and artificial intelligence models in a unified manner to provide artificial intelligence model services in the e-government external network environment to subordinate units or regions. Cities should carry out deployment applications under the unified requirements of provinces (autonomous regions, municipalities directly under the central government), and county-level and below should, in principle, reuse the intelligent computing power and model resources of superiors to carry out applications and services, and no longer independently carry out the construction and deployment of government models.
（3） Explore the implementation of centralized management and reuse
Government departments should explore the construction of an intensive deployment model of "one place construction, multiple places and multiple departments reuse", coordinate and promote the deployment and application of government models, and prevent the formation of "model islands". Provinces (autonomous regions, municipalities directly under the central government) should establish a unified service platform for artificial intelligence big models in the field of government affairs, and integrate and jointly build it with government cloud management platforms, government application and component management platforms, etc., to integrate the intelligent computing power, government big models, government datasets and other resources of the e-government external network in the region into unified management, form a "one account" of element resources, support the operation and monitoring of government big models, provide resource application and scheduling services, and promote efficient reuse. The national industry regulatory authorities shall explore the unified training and construction of vertical government models in segmented fields according to business needs and development needs, strengthen collaborative deployment with provinces (autonomous regions, municipalities directly under the central government), and deepen the intelligent empowerment of industry fields across levels and regions. Vertical management departments should strengthen the coordinated deployment and management of resources such as models, computing power, and data to avoid resource waste.
（4） Continuously strengthen the data foundation
Government departments should strengthen government data governance, continuously improve data quality, accelerate the construction of high-quality government data sets and knowledge bases that objectively reflect public policies, institutional norms, business processes, and governance effectiveness, and support the optimization and training of government big models. The classification and grading management of the government affairs big model involves data, strengthening the management of training data, fine-tuning data, knowledge base, etc., establishing a ledger and recording detailed information such as data sources, types, and scales to ensure reliable and traceable data sources, accurate and effective content. Relying on the government data sharing and coordination mechanism, we will coordinate the results of data governance, promote the co construction and sharing of high-quality government data sets, and collect and manage the generated data. Explore the governance path of government knowledge based on big models, build a trustworthy knowledge base, and ensure the authority, accuracy, and timeliness of data sources.
4、 Operation management
Government departments should strengthen the operation and management of artificial intelligence big models in the field of government affairs, improve management systems, operation modes, and security requirements, and orderly promote the deployment and application of artificial intelligence big model technology, products, and services in the field of government affairs.
（1） Clearly define application management requirements
Government departments should coordinate the reduction and empowerment of burdens, strictly implement relevant requirements such as the "Several Provisions on Rectifying Formalism to Reduce Burden at Grassroots" and the "Several Opinions on Preventing and Controlling Formalism at Fingertips", avoid blindly pursuing technological leadership and conceptual innovation, avoid redundant and ineffective construction, avoid building without review, avoid forced or ineffective use, avoid data collection and repeated requests, and effectively prevent "digital formalism". The deployment and application of artificial intelligence models in the field of central and state government affairs should be included in the overall planning of national government informationization. Government departments should establish and improve a comprehensive management system for the deployment and application of artificial intelligence big models in the field of government affairs, clarify application methods and boundaries, implement the "auxiliary" positioning of artificial intelligence big models, and timely solve new problems that arise during deployment and application. Risk warnings should be prominently displayed on the application interface of the government model, clearly indicating the limitations of the model services and properly labeling the output content of the model. For the application scenarios of artificial intelligence big models that represent government departments providing services to the public and enterprises, such as intelligent question answering and auxiliary processing, the content review system process should be strictly implemented. Based on the characteristics of the scenario and technical capabilities, measures such as manual review, real-time risk control of content generation, and cross validation of multiple models should be reasonably adopted to prevent risks such as model "illusion", ensure that the output content does not exceed the business scope, guarantee content accuracy, and maintain the credibility of government departments.
（2） Continuously promote iterative optimization
Government departments should regard continuous iterative optimization as a key link in the deployment and application of artificial intelligence models, establish a regular update mechanism, accelerate functional optimization, and deepen scenario applications. Pay close attention to technological developments and continuously update and optimize the basic models and security capabilities of artificial intelligence models in the field of government affairs. Establish an efficient data collection and processing mechanism, update the input data and knowledge base that support the operation of artificial intelligence models in a timely manner, clean and annotate them in a timely manner, supplement and optimize the training dataset, and continuously improve the model's capabilities. Establish a user evaluation and feedback mechanism for artificial intelligence models in the field of government affairs, collect and process user needs in a timely manner, and drive iterative optimization with user feedback.
（3） Solidly carry out safety management
Government departments should establish a security responsibility system, clarify the security responsibilities and tasks of participating entities in data processing, large-scale model training, and scenario application stages, and do a good job in user identity recognition and permission management. When providing artificial intelligence big model services, government departments should comply with relevant regulations such as the Interim Measures for the Management of Generative Artificial Intelligence Services, use data and basic models from legitimate sources, fulfill obligations such as algorithm filing and security assessment in accordance with the law, sign service agreements with users, and clarify the rights and obligations of both parties. Build a classification and hierarchical governance system for artificial intelligence models in the field of government affairs, improve security management processes, and develop emergency response plans for potential security risks. Carry out detection and disposal of adversarial attacks on government models, identify and intercept prompt word injection, resource consumption attacks, etc. Strengthen the content security management of government affairs models, comprehensively utilize semantic recognition, rule libraries, model algorithms, etc., to identify, analyze, and control various modal input and output contents, establish reasonable response and rejection mechanisms, and timely discover and dispose of illegal and harmful information, sensitive content, etc. Give full play to the advantages of news media content review, do a good job in content review and control of government model training data, and strengthen the monitoring and management of government model content. Manage the operation logs of the government model application and conduct regular audits of the log records. Promote the formation of a security risk threat information sharing and emergency response mechanism, timely handle and report security incidents in accordance with regulations, and enhance the ability of artificial intelligence to respond to security risks.
（4） Strictly implement confidentiality requirements
Government departments should strengthen data security, confidentiality, and personal information protection in the process of model training, deployment, and application. They should adhere to bottom line thinking, strictly implement confidentiality discipline requirements such as "no internet access for classified information, no internet access for classified information", and take measures such as installing confidentiality "guardrails" to prevent state secrets, work secrets, and sensitive information from entering non classified artificial intelligence models, and prevent the risk of leakage caused by the aggregation and association of sensitive data. Develop and improve the confidentiality management system related to the application of artificial intelligence big models in the field of government affairs, standardize the confidentiality management of the entire process of selecting, deploying, training, using, and abolishing artificial intelligence big models. The application of artificial intelligence models in classified information systems is being steadily promoted in accordance with the requirements of the national secrecy administrative management department.
5、 Guarantee measures
（1） Strengthen organizational implementation
Strengthen overall planning and coordination, steadily and orderly promote the standardized application of artificial intelligence big models in government services, social governance, office work, auxiliary decision-making and other fields. Accelerate the construction of the national standard system for artificial intelligence models in the field of government affairs and the development of key standards, clarify the work norms for application effect evaluation, system technical requirements, intelligent technology application, etc., and support the deployment of applications to achieve practical results. Timely summarize and promote typical scenarios and innovative applications of deploying and applying artificial intelligence models in the field of government affairs, and promote reuse and efficiency improvement. Strengthen the funding guarantee for the deployment and application of artificial intelligence big models in the field of government affairs, introduce market-oriented product and service competition mechanisms, explore the operation mode of enterprise construction and operation, government purchase of services, and settlement of fees based on usage, and create an efficient and sustainable ecosystem of government big models.
（2） Carry out monitoring and evaluation
Build a comprehensive monitoring and evaluation system for the deployment and application of artificial intelligence models in the field of government affairs, and carry out monitoring and evaluation work in a timely manner. Establish a security evaluation mechanism for the government affairs big model, fully test and verify the model algorithm, generated content, application functions, configuration environment, linked data, vulnerability risks, etc. before going online, and rectify and strengthen the problems and hidden dangers discovered. Strengthen the real-time monitoring and analysis of the operation status, response time, accuracy, security, and potential risks of artificial intelligence large-scale model systems in the field of government affairs, promptly identify problems, and take effective measures to solve them. Conduct effective evaluation of the application effectiveness of artificial intelligence models, summarize experience in a timely manner, continuously iterate and optimize, and promote the deployment of applications to achieve practical results.
（3） Do a good job in training and publicity
Develop a training curriculum system covering the theory, technology, application, safety, ethics, industry, and other aspects of artificial intelligence big models, conduct training on artificial intelligence literacy and skills, enhance the cognitive level of leaders and cadres towards artificial intelligence, and strengthen the application ability and level of staff. To carry out publicity and education for the public, enhance the digital literacy of the entire population, actively respond to user concerns, correctly guide society's understanding and expectations of the applicable population, scenarios, and uses of artificial intelligence models in the field of government affairs, objectively reflect the role of artificial intelligence models in optimizing government services, meeting the needs of the public and enterprises, and improving social governance levels.
