ChinaTech #20 | Another week, another AI regulation
... Encouraging AI adoption by the Government Sector
A few days ago, the Cyberspace Administration of China (CAC) released the “Guidelines for the Deployment and Application of Large-Scale Artificial Intelligence (AI) Models in Government Affairs”.
This document encourages the use of AI by government departments in China. Several application scenarios have been laid out and range from admin tasks, service delivery and law enforcement to better decision making.
(1) Government Services
1. Intelligent Q&A. This service integrates data from regional, departmental, and field-specific business resources and knowledge bases, leveraging technologies such as natural language understanding, search-enhanced generation, and knowledge graphs to provide convenient online government consultation services. This service enhances accurate understanding of public concerns, generates reference answers in real time, helps resolve public concerns, and improves ease of access to information.
2. Assisted processing. By integrating data such as government service guides, frequently asked questions, user reviews, and historical processing records, and leveraging technologies such as intelligent matching and automated processing, this service provides one-stop government service assistance, including intelligent guidance, personalized guidance, pre-filled forms, assisted review, progress inquiry, and reminders. This assists staff in efficiently reviewing materials and facilitates the public and businesses in handling matters.
3. Direct and fast access to policy services. Build a policy service knowledge base, detailing policy requirements, policy labels, push conditions, application and redemption processes, and other related content. Utilize the “policy finds people” and “policy finds businesses” algorithmic models to strengthen analysis of public and business needs, achieve intelligent policy matching, and promote proactive and precise delivery and one-stop processing of services that benefit both the public and businesses.
(2) Social Governance
4. Intelligent monitoring and inspection. Using drones, video surveillance, smart sensors, and other equipment, along with computer vision and other technologies, real-time analysis of surveillance videos, images, and IoT sensor data is performed to assist personnel in real-time monitoring of infrastructure such as buildings, roads, gas, bridges, water supply, drainage, heating, and integrated pipeline corridors. This allows for the timely detection of abnormal behavior, environmental issues, or facility failures. Potential risks and hidden dangers are automatically identified, prompting timely alerts and recommendations based on the severity of the abnormality, improving monitoring and inspection efficiency.
5. Assisting law enforcement and supervision. Using technologies such as voice recognition, video analysis, knowledge graphs, and logical reasoning, the system assists law enforcement personnel in entering case information into the system in real time, identifying clues to problems, generating case reports, quickly retrieving legal basis and judicial interpretations, and searching for similar typical cases. It also provides targeted case handling advice, improving the efficiency and standardization of law enforcement and supervision.
6. Market Risk Forecasting. Using generative time series analysis models and anomaly detection algorithms, we monitor and conduct in-depth analysis of various market data, capturing market trends, including fluctuations in economic indicators and anomalies. We predict potential market risks, assess their impact on the economy and society, and assess economic trends. We also issue timely warnings to support government management and social governance.
(3) Office work
7. Assisting with document drafting. Leveraging the generative capabilities of large language models, by building a local knowledge base and pre-set templates, this service provides staff with writing suggestions, assists in drafting documents, and checks, proofreads, and optimizes formatting and content, improving work efficiency and reducing the burden on grassroots staff.
8. Data Retrieval. Leveraging knowledge graph construction and information retrieval technologies, we accurately understand staff members’ data retrieval needs and enable rapid retrieval, precise location, multi-dimensional sorting, intelligent association, and comparative analysis of government information, helping staff improve data retrieval efficiency and accuracy.
9. Intelligent task assignment. Leveraging technologies such as natural language understanding and multimodal recognition, we build multi-dimensional task classification and assignment rules. This automatically identifies, accurately categorizes, assists with filling out, and prioritizes tasks such as incoming letters, calls, and work orders. This enables assisted distribution and intelligent dispatching, improving task assignment efficiency.
(IV) Decision-making assistance
10. Disaster Early Warning. This system conducts big data correlation and comprehensive analysis of multi-source, multi-dimensional, and multi-modal data from satellites, ground sensors, geological monitoring stations, forecasts, early warnings, and disaster risk surveys. This system identifies abnormal fluctuations, predicts potential natural disasters, and issues early warnings to assist government departments in taking timely and effective measures to mitigate disaster risks and reduce losses.
11. Emergency Response. Leveraging technologies such as reinforcement learning, analyze and assess the nature, characteristics, degree of harm, scope of impact, development trends, and public reaction to public safety and other emergencies, promptly identify and warn of potential risks. Rapidly simulate the effectiveness of emergency response plans based on emergency scenarios and resource distribution, provide scientifically sound emergency response recommendations, optimize rescue resource allocation, and improve the speed and efficiency of emergency responses.
12. Policy Evaluation. Leveraging the inference and analysis capabilities of large AI models and data mining, analyze public feedback, market reactions, economic indicators, and social satisfaction, construct multi-dimensional indicators, and assess the degree of achievement of policy objectives, policy impact, and potential problems, supporting policymakers in optimizing policies.
13. Intelligent-Assisted Review. Leveraging self-learning generalized cognition, human-like review reasoning, and multimodal intelligent analysis capabilities, project reviews are conducted against relevant requirements. Project documents are deeply scanned and intelligently analyzed, providing review opinions and suggestions to help improve the efficiency and scientific nature of project reviews.
Further, the document lays great emphasis on standardizing operations across departments and avoiding fragmentation and duplication of efforts. Government departments have been called upon to explore a model of “single-site construction, multi-location, and multi-department reuse”.
Leveraging the “East-West Computing” initiative and the nationwide integrated computing network, they should comprehensively promote the layout of intelligent computing infrastructure and implement centralized security management and systematic technical protection measures to avoid the security risks of “fragmentation.”
Provinces (autonomous regions, and municipalities) should establish a unified service platform for AI big models in the government sector, integrating it with government cloud management platforms, government application and component management platforms, and other platforms. This platform will bring resources such as e-government extranet intelligent computing power, government big models, and government data sets within the region under unified management, creating a “single account” of key resources. This will support the operational monitoring of government big models, provide resource application and scheduling services, and promote efficient reuse.
The document talks about the need to adopt the right “implementation path” and stresses, yet again as so many recent documents do, on building the foundational data ecosystem to prop all this up.
For scenarios with high versatility and rich data resources, such as intelligent question-and-answering and document drafting assistance, mature model products and services on the market that have been registered with the cybersecurity and informatization authorities should be adopted. For scenarios with high professional expertise and complex business logic, such as assisting law enforcement and market risk forecasting, targeted training can be conducted using domain expert knowledge and professional data to create vertical models. While ensuring security and preventing the disclosure of state secrets, work secrets, and sensitive information, full use should be made of internet computing power and model resources to deploy and apply large-scale AI models in government affairs. Exploration of innovative applications such as government intelligence and embodied intelligence is encouraged.
What’s interesting is this section warning against over-engineering.
They (government departments) should avoid blindly pursuing technological advancement and conceptual innovation, avoid duplicative and ineffective construction, avoid building without prior review or neglecting construction, avoid forced and ineffective use, avoid multiple data collection and repeated requests, and effectively guard against “digital formalism.”
As a way of implementation, government departments have been asked to iteratively experiment, establish a safety and confidentiality framework, and carry out staff training and publicity for an efficient rollout.
Only a few weeks after the China AI+ initiative, these documents encouraging the use of AI by different sectors of the Chinese economy are important and necessary to track.