Korea’s first K-AI City pilots are being designed to let artificial intelligence learn across urban systems and support city operations. But this ambition raises a governance question even before any algorithm begins making recommendations. AI can only interpret the city as it appears in its data, and that picture is often incomplete, uneven, or missing certain communities altogether. As machine judgment moves closer to public decision-making, cities must understand not just what the system sees, but also what it overlooks—and who ultimately remains responsible for the outcomes.
Korea’s K-AI City Push Moves AI Closer to Urban Decisions
South Korea is beginning to give the concept of an AI city a concrete policy form. On June 18, the Ministry of Land, Infrastructure and Transport, or MOLIT, selected Wonju and the joint Cheonan-Asan area as the country’s first AI-specialized pilot cities. MOLIT described the initiative as a K-AI City model where AI infrastructure can learn from data collected across the city to support safer and more efficient urban operations.
That ambition goes deeper than adding an AI chatbot to city hall. In May, the Korea Research Institute for Human Settlements, or KRIHS, argued that AI is emerging as a factor capable of reshaping urban operating systems themselves. Its policy brief identified agentic AI as a potential actor capable of autonomous judgment and execution within urban operations.
This is where Dr. Sung Jin Park sees a fundamental change in the smart city discussion. Park is Founder and CEO of SPARKCITY and has worked across smart city consulting, product development, and business strategy at Samsung SDS and LG U+, alongside academic work on urban AI and technology governance.
Smart cities have generally used connectivity, integrated data, monitoring, and automation to improve urban efficiency, Park told KoreaTechDesk as discussion on smart city startup expansion continues. AI cities introduce a more consequential question once software begins interpreting conditions across systems and influencing decisions.
“How much autonomy should we give AI in operating our cities, and how should that autonomy be governed?” The answer begins before AI reaches the decision layer. Because it actually starts with the information that allows the system to perceive the city at all.
AI Does Not See a City Directly. It Sees the Data That Represents It
Urban data is continuously generated through public services, administrative records, sensors, infrastructure, transportation systems, institutions, and human activity. An AI system tasked with understanding urban conditions depends on those records as its representation of what is happening.
Park cautioned that this representation can be incomplete, inaccurate, biased, distorted, or unavailable. Some communities may appear less frequently in administrative datasets, while other groups, locations, or activities may produce much richer digital traces.
That creates an important distinction for AI city governance. A technically sophisticated model can still reason across an incomplete picture of urban reality.
“If they are invisible in the data, they may also become invisible to AI-driven urban decisions,” Park said. And her warning has really identified a governance risk that becomes more important as AI gains greater influence over how cities interpret needs, allocate attention, and support administrative decisions.
International public-sector research points to the same underlying vulnerability. The OECD’s 2025 analysis of 200 government AI use cases warned that skewed data can contribute to harmful decisions, while weaknesses in data access and representation can widen digital divides and propagate errors. The report found that 45% of the examined use cases supported decision-making, sense-making, or forecasting, placing data quality directly upstream of government judgment.
Korea Is Turning Public AI Governance Into an Institutional Responsibility
The timing of Park’s argument is particularly relevant because Korea is about to change how public institutions govern AI.
The revised Act on the Promotion of Artificial Intelligence and Data-Based Administration was scheduled to take effect on August 28, 2026. The law covers central government bodies, local governments, public institutions, local public corporations, and other designated entities. It formally expands Korea’s previous data-based administration framework to include AI-based policymaking and decision support.
Several provisions address the institutional layer behind those systems. Public bodies must appoint an AI and Data-Based Administration Officer responsible for coordinating AI and data policy, data sharing, training-data management, trustworthiness, and risk management.
The law also requires public institutions to take measures to maintain an appropriate quality level for training data intended for inter-agency sharing and use. More broadly, public bodies are expected to maintain the recency, accuracy, and interoperability of their data.
These requirements matter for AI cities because responsibility can no longer sit only with the team purchasing or developing an individual application. Once AI becomes embedded across urban functions, data quality and risk management become continuing administrative responsibilities.
Park argues that this requires permanent institutional capability inside city governments, rather than oversight that ends with an individual AI project.
“Cities need professionals who can identify data blind spots, monitor data quality and consistency, examine how AI systems interpret urban data, audit AI-supported decisions, and evaluate their real-world outcomes.”
Seoul Shows How Quickly Municipal AI Can Spread Across an Organization
Seoul already offers a glimpse of the organizational scale involved.
In March, the Seoul Metropolitan Government launched a 12-member AI Committee covering policy, technology, industry, and ethics. Its 2026 AI administration plan includes 61 projects involving 17 city offices and affiliated organizations, covering several categories of AI capability.
Seoul had already introduced municipal AI ethics guidelines in January based on five principles: public interest, fairness, transparency, accountability, and safety. The guidelines specify that human supervision and responsibility should remain throughout AI use, with final judgment and responsibility resting with the people administering the service.
That principle becomes harder to operationalize as AI spreads across dozens of functions. Human oversight has little meaning if nobody has a clear view of which systems are operating, which datasets they use, how recommendations move through administrative workflows, and where intervention is possible.
Korea’s revised public AI law addresses part of this visibility problem. Public institutions will have to maintain information on the type, purpose, and data associated with their AI-enabled services and submit those records to the Ministry of the Interior and Safety. The ministry must maintain the resulting inventory and publish the list annually.
A city cannot govern AI systems it has not first systematically identified.
Public AI Adoption Is Moving Faster Than Outcome Monitoring
The institutional challenge extends well beyond Korea.
The OECD’s Digital Government Outlook 2026 found that 35 of 36 OECD countries already use AI in at least one area of government. Thirty countries have at least one institution responsible for governing public-sector AI.
The monitoring layer remains much thinner. Only 10 of 36 OECD countries reported conducting financial or non-financial impact measurement of government AI use cases, while 11 reported post-deployment audits. Only eight had citizen feedback or complaint mechanisms related to government AI use.
That gap matters for cities because responsible AI cannot simply end when a system passes procurement, testing, or technical evaluation. The consequences emerge during operation, when AI encounters changing populations, new data, exceptional events, and administrative conditions that were difficult to reproduce before deployment.
Park’s concept of permanent oversight therefore reaches beyond conventional model monitoring,
“For this reason, I believe cities need permanent organizational capabilities, not temporary AI task forces, to continuously govern, monitor, and evaluate the entire data-to-decision chain.”
Seoul Residents Are Already Asking Who Makes the Final Call
Public attitudes suggest that institutional accountability is not an abstract policy concern.
Before establishing its AI Committee, Seoul conducted an online survey between February 27 and March 11 with 9,425 participants. Among respondents, 60.7% said clear responsibility and final human review were more important than processing speed when AI is introduced into public administration. Another 57% preferred sufficient verification before innovative technologies are deployed.
And those findings reveal a useful tension as cities seek faster and more proactive services.
People can value AI-enabled convenience while still expecting an identifiable person or institution to remain accountable when machine-supported judgment affects public administration.
That expectation becomes increasingly important as AI moves beyond simple assistance. Systems can summarize conditions, forecast risks, recommend actions, prioritize cases, coordinate information, and potentially initiate limited actions under predefined rules.
Each additional layer of autonomy increases the importance of knowing where machine judgment ends and accountable public authority begins.
An AI City Still Depends on the Digital City Beneath It
Park does not see AI cities as replacing the smart city concept. Instead, she views them as an evolution built on the digital foundation smart cities have spent years developing.
“The foundation of an AI City is data.”
Digital infrastructure creates urban data. Sensors and public systems continuously generate information. Communication networks move it. Interoperable platforms allow information to be combined, and AI can then interpret those representations to support urban decisions.
Korea’s first K-AI City program reflects that dependency. MOLIT’s model explicitly links AI capability with infrastructure that allows systems to learn from data collected across the urban environment.
This means the intelligence of an AI city cannot be judged only by the sophistication of its models. Its performance also depends on the quality of the underlying urban information, the institutional capacity managing that information, and the human accountability surrounding decisions built on it.
The Hardest AI City Problem May Be What the System Fails to Notice
Cities have spent years asking how digital technology can make urban systems more responsive. AI adds another possibility: software that can interpret complex conditions and support decisions at a scale difficult for human teams to process alone.
But greater analytical capability can create a subtle risk. The more confidence institutions place in machine interpretation, the more consequential its blind spots become.
Park believes this is why AI city governance needs continuing institutional scrutiny rather than a one-time approval process.
“Before asking how much AI cities should adopt, city governments should first ask whether they have the organizational capacity to continuously monitor what their AI systems see, learn, decide, and ultimately fail to see.”
Korea is beginning to build that institutional layer through dedicated AI responsibility, training-data obligations, service inventories, municipal ethics rules, and governance bodies. The harder test will come when those mechanisms have to operate continuously across real urban systems.
An AI city may eventually become very good at identifying patterns that humans cannot see. Its credibility will depend just as much on the city’s ability to recognize what the algorithm cannot see, and to keep responsibility visible when machine judgment enters public decisions.
Key Takeaway
- Korea’s AI city governance is shifting beyond model performance. MOLIT’s K-AI City pilots in Wonju and Cheonan-Asan tie AI learning directly to urban data, making data quality and accountability core issues.
- Urban AI only works with the data it has. Gaps or bias in datasets can hide people, places, or conditions from AI-driven decisions.
- Continuous oversight is now required. Cities must constantly monitor data quality, AI outputs, decisions, and blind spots—not just rely on one-off project teams.
- Korea is institutionalizing AI responsibility. The revised AI and Data-Based Administration Act (effective Aug 28, 2026) mandates AI/data officers, training-data governance, and AI service inventories in public institutions.
- Seoul is scaling AI governance across government. It is running 61 AI projects across 17 agencies, backed by ethics rules that keep human responsibility and oversight central.
- Citizens prioritize accountability over speed. In Seoul’s 2026 survey of 9,425 people, 60.7% preferred clear human responsibility and final review over faster AI processing.
- Global governance still lags behind adoption. While AI use in government is widespread, few countries consistently conduct impact assessments, post-deployment audits, or collect citizen feedback.
- The core AI city challenge is institutional. As AI shapes urban decisions, cities must clearly define what the system sees, what it misses, and who is accountable for outcomes.
---
Autor(en)/Author(s): Richard Park
Dieser Artikel ist neu veröffentlicht von / This article is republished from: Korea Tech Desk, 12.09.2026
Bitte besuchen Sie/Please visit:
