AI opens the future of government public services As the wave of digital transformation sweeps across society, the adoption of artificial intelligence (AI) in government services is emerging as a critical issue. The recent trial of the 'GOV.UK Chatbot' operated by the UK government's digital service organization stands out as a prime example of this change. This project represents the first step in improving public service operations with AI systems, aiming to make communication between the government and its citizens more efficient and seamless. Examining the successes and challenges of such experiments can provide valuable lessons for the Korean government as it designs the future of digital administration. A report published on the UK's GOV.UK blog, titled '5 things we learned from testing the GOV.UK chatbot: an AI assistant for government,' offers concrete insights based on real data. According to this report, the chatbot's accuracy increased from an initial 76% to 90% over two pilot operations. This improvement is primarily attributed to the continuous efforts of data scientists and the advanced learning of the AI model. The project has received positive evaluations, particularly for its ability to quickly rectify issues discovered during the initial operation and systematically enhance system performance. In the first pilot, the UK government tested the chatbot's basic performance, collecting data on the types of questions users asked and what constituted a satisfactory answer. At this stage, an accuracy of 76% was deemed insufficient for government service standards. However, in the second pilot, the collected data was analyzed to retrain the AI model, and frequently occurring error patterns were identified and corrected. As a result, accuracy improved to 90%, demonstrating the critical importance of iterative testing and improvement when introducing AI into government services. However, this test was not merely about technical achievements. There were a total of 508 'jailbreak' attempts during the test, vividly illustrating the paramount importance of security for public service AI. A jailbreak attempt refers to an effort to induce an AI chatbot to provide inappropriate or harmful responses unintentionally. The GOV.UK chatbot successfully defended against all these attacks through multi-layered safeguards built into the system. This suggests that when adopting AI for government services, thorough preparation for security and safety is essential, not just technical performance. The five key lessons presented in the GOV.UK blog report are as follows: First, AI chatbot accuracy can be improved through continuous data collection and model enhancement. Second, security mechanisms must be embedded into the system from the design stage. Third, systematically analyzing user experience data is crucial for service improvement. Fourth, finding a balance between response speed and accuracy is important. Fifth, given the nature of government services, transparency and trustworthiness must be prioritized above all else. The most noteworthy aspect of this case is that it simultaneously revealed both the potential and limitations of AI. While many governments focus on the economic benefits of technology adoption, there is also a strict demand for verification of how AI systems handle citizens' personal information and how effectively they can provide services. For instance, the GOV.UK chatbot's average response time was measured at approximately 10.7 seconds. While this might be considered slow for users expecting real-time conversations, it is also the necessary processing time to provide accurate and reliable government information. The UK government continues to monitor the impact of this response time on user satisfaction, seeking ways to improve speed without sacrificing accuracy. Lessons from chatbot adoption success and failure Regarding response time, the GOV.UK team is accumulating user experience data to refine its improvement strategy. Generally, online service users expect a response within 3 seconds, but the UK government believes that for government services, accuracy should take precedence over speed. Therefore, a response time of 10.7 seconds is currently considered acceptable, with plans for gradual improvement through technological advancements and system optimization. Similar issues and potential for development can be found in Korea's public services. The Korean government is already applying AI in various public service sectors, accelerating the digitalization of administration. Chatbot services are being piloted on platforms like Government24 and Minwon24, assisting citizens with administrative tasks. However, there are also criticisms that more effort is needed to achieve service completeness and secure citizen trust. In particular, technical limitations still exist for AI to accurately understand and process the linguistic characteristics of Korean, complex administrative terminology, and di
Related Articles