The era of public officials using artificial intelligence (AI) to directly develop administrative programs has arrived. A senior official at the Sejong City Disaster Safety Situation Room created a disaster response platform using generative AI. A task that previously took 10 minutes—classifying disaster types, selecting responsible departments, and drafting situation reports—is now completed as a draft by AI in 10 seconds. The official then reviews and sends it. In critical moments, 10 minutes can determine someone’s life. This is a welcome change.
The government is actively supporting public officials’ use of AI. On September 8, a Cabinet meeting reported measures to promote AI development by public officials. The government plans to train 20,000 AI-capable public officials by 2030. Experiments following established procedures will be protected under the active administration immunity system unless intentional or gross negligence is proven. Within two months of a pilot program, 770 projects were registered.
The direction is correct. Frontline public officials best understand on-the-ground issues. If the time previously spent on repetitive data organization and report writing is redirected to caring for elderly individuals living alone or inspecting disaster sites, citizens will feel the change. However, when AI impacts citizens’ lives, the issue becomes more complex. Suppose a resident hears at a community center, “The AI analysis shows you are not eligible for welfare.” If this is due to a single incorrect income data entry, to whom should they appeal? No citizen will accept the explanation, “This is what the AI did.”
The Sejong City case is commendable because a person ultimately reviews the results. The question is whether this principle will be maintained during expansion. When a public program is adopted by another agency, the review process might be skipped. If AI judges flood risk as low and the official follows this judgment, who is responsible—the official who created the program, the one who used it, the approving agency, or the adopting agency? AI does not apologize, face disciplinary action, or compensate for damages. However, if all responsibility is placed on the developing official, no one will attempt innovation. While protecting public officials’ experiments, the government’s responsibility to citizens must not be exempted.
Institutional safeguards do exist. The AI Basic Act requires businesses providing AI that significantly impacts citizens’ lives or rights to prepare for such risks. The Personal Information Protection Act also allows individuals to request explanations for automated decisions made by AI or to reject such decisions in certain cases. However, if public officials merely perform a perfunctory review of AI results, can this be considered human judgment? If responsible review becomes a mere approval process, gaps emerge between AI errors and administrative accountability.
Three measures are needed. First, even during pilot operations, services affecting citizens’ rights must be managed as formal services. Second, if another agency’s program is used, the adopting agency must directly verify and take responsibility for the results. Third, if AI was involved in the decision, the official must inform the affected party, even after final review. The party should also have the right to request a re-examination by the official.
The level of an AI government is not determined by the number of programs or processing speed. Who explains, corrects, and takes responsibility when AI makes a mistake determines the government’s true level. Public officials should be given freedom to experiment, but citizens must be guaranteed a government where humans ultimately take responsibility.
Leave a comment