Generative AI holds transformative potential for public administration by drastically reducing task processing times through automation of repetitive tasks and support for complex decision-making. In areas such as large-scale administrative data processing, civil complaint handling, and policy analysis, generative AI can streamline operations and improve service delivery. Particularly, when paired with a collaborative workflow where AI drafts documents, summarizes meetings, analyzes data, or reviews legal texts, and public officials provide expert oversight both the quality and efficiency of administrative services can be significantly enhanced.
Furthermore, the introduction of AI-powered civil consultation systems capable of operating 24/7 can elevate citizen satisfaction while easing the workload of public servants. These innovations go beyond operational efficiency, driving qualitative advancements in public services and enhancing the precision of policymaking ultimately contributing to the overall competitiveness of public administration.
Globally, leading nations such as the United States, the United Kingdom, and Singapore have already adopted generative AI across a range of administrative functions. Singapore’s "AI Singapore" initiative has notably reduced processing times by embedding AI throughout its public services. The UK has improved public satisfaction through AI service platforms developed by the Government Digital Service (GDS). Meanwhile, digital pioneers like Estonia have fully implemented AI-powered e-government systems that enable both process automation and personalized services. These international developments signal not just a technological upgrade but a paradigm shift in governance, underscoring the necessity of generative AI innovation for maintaining national competitiveness.
However, several critical limitations must be addressed when deploying GPT models in administrative contexts. Most notably, GPT can produce plausible yet inaccurate outputs a phenomenon known as "hallucination" which makes human verification indispensable, especially in high-stakes decisions. Moreover, the use of sensitive administrative data poses security risks, including the potential transmission of confidential information to external servers.
GPT models are also constrained by the scope and currency of their training data, limiting their ability to reflect recent legislative or institutional changes. They may struggle with complex issues that require nuanced interpretation or ethical reasoning. Overreliance on such tools can weaken the expertise of civil servants, and the opacity of the underlying algorithms may hinder transparency and accountability in public decision-making.
To overcome these challenges, Retrieval-Augmented Generation (RAG) has emerged as a powerful solution. RAG enhances response accuracy by retrieving relevant information from trusted databases or documents before generating a response. This is especially valuable in public administration, where precision is paramount. By integrating real-time access to current laws, regulations, and institutional guidelines, RAG ensures that AI-generated responses remain fact-based and up to date.
This approach dramatically reduces hallucination errors and increases the reliability of AI-generated content, empowering public officials to use such tools with greater confidence. The result is improved administrative efficiency and more informed decision-making.
In Seoul, a vast archive of disaster-related documents—including white papers, incident reports, response manuals, crisis management protocols, and case studies has been systematically digitized. Utilizing state-of-the-art embedding technologies, the core information from these documents has been vectorized and stored in a specialized vector database.
This vector database serves as the foundation for Seoul’s RAG-based generative AI system, enabling it to rapidly retrieve relevant, context-specific information during disaster events and provide optimized response strategies. The system supports fast, accurate decision-making and underpins an intelligent disaster management framework capable of delivering real-time guidance to on-site emergency response teams.
To rigorously evaluate the effectiveness of Seoul’s generative AI system in the field of disaster and safety management, a comprehensive performance assessment was conducted. A quantitative survey was carried out among public officials from key departments such as the Disaster and Safety Policy Division, the Disaster and Safety Prevention Division, and the Construction Innovation Office focused on five criteria: response specificity, usefulness, retrieval speed, reliability, and length.
In parallel, a panel of subject-matter experts in disaster response, safety, and disaster prevention engineering conducted a qualitative evaluation. Through in-depth interviews and case analyses, the panel affirmed the reliability of the system and its practical relevance. The experts especially highlighted the system’s ability to perform multidimensional analysis and recommend actions based on historical precedents, recognizing its value as a field-ready decision-support tool.