ROUNDTABLE Korea needs AI-ready ontology, should not end with developing models

Korea needs to diversify its current artificial intelligence (AI) strategy which is largely confined to developing advanced models, and expand its efforts to build AI-ready knowledge systems, infrastructure and a regulatory environment that supports practical applications, experts said during The Korea Times roundtable, “Beyond US-China AI race: Korea's strategic edge.”

The participants called for greater efforts to build ontologies — structured knowledge systems that organize vast amounts of government data, institutional knowledge and the relationships between them into forms AI systems can understand and use. They also agreed to the necessity of creating a sandbox area where AI-powered technologies could be tested and deployed under a more flexible regulatory framework.

Reform Party of Korea Rep. Lee Jun-seok stressed that the government’s AI strategy should focus on building national ontologies, not distributing free chatbots to the general public.

“If the government's goal was to drive the nation's AI transformation, a key policy should have been to require each ministry to develop its own ontology and submit a detailed implementation plan,” Lee said.

“Instead, the government allocated budgets without considering such a strategy, leaving officials unsure of what they should actually do. As a result, the only answer they can come up with is to buy more graphics processing units.”

During last month’s National Assembly audit of the Public Procurement Service, Lee’s office converted raw data on 1.05 million products registered with the agency and 2.8 million procurement orders into ontologies and developed an AI program to analyze the data.

The analysis uncovered signs of potential misuse of 562.8 billion won ($418.4 million) in taxpayer money, including cases in which power strips normally priced at 60,000 won were supplied in large quantities to public institutions for 200,000 won each.

“We are now working on another project to convert all of Korea’s laws, enforcement decrees and municipal ordinances into ontologies, and have already identified more than 10 broken links between legal provisions, including cases where a law stipulates fines under provisions that no longer exist,” he said.

“It is important to deploy large numbers of software engineers and architects to government agencies and other specialized fields where they can work on building these ontologies.”

Lim Jong-in, former presidential cybersecurity adviser and Soongsil University AI committee chair, noted that Korea should continue developing its own foundation models, even if they struggle to compete with leading models, because the country should not fully rely on foreign-developed AI models in sensitive areas such as national security, chip designing and cybersecurity.

“Samsung Electronics uses OpenAI's and Anthropic's Claude models for its operations, but there is a reason it is also using its own small AI model, Gauss, to prevent sensitive design assets and specifications from being exposed,” Lim said. “We need our own AI models for economic security, national security and the protection of personal information.”

He also noted that the adoption of global cybersecurity platforms such as Palantir and CrowdStrike is becoming increasingly common among large businesses, but many small and medium-sized enterprises and public institutions cannot afford to use them. He therefore stressed the need for Korea to continue developing affordable domestic AI models specialized in cybersecurity.

Purpose-built AI city

To enhance Korea’s competitiveness in the AI era, Lee proposed a purpose-built AI city in Saemangeum, a massive reclaimed area in North Jeolla Province. He said the city would serve as a large-scale test bed for about 500,000 residents who voluntarily agree to live under significantly relaxed regulations, eliminating the inefficiencies of trying to integrate AI technologies into existing communities.

“For example, using humanoid robots for food delivery would be highly inefficient,” Lee said. “If a robot tried to board a crowded elevator during lunchtime, people would probably push it out. But in an AI-native city, we could design buildings with small elevators dedicated to deliveries or incorporate drone landing pads from the beginning.”

Developing AI-powered autonomous vehicles and robots does not necessarily mean that every capability must be built into the systems from scratch. In the early stages of connected vehicle development, urban sensors can help enhance their capabilities, while factories operating large numbers of humanoid robots also need environments where sensors installed throughout the facilities can assist the robots' movements.

Lee said Korea lacks such infrastructure, citing an example from Dongtan, his constituency in Gyeonggi Province, where a Chinese autonomous vehicle company expressed interest in testing its technology there as a model city but faced restrictions that prevented autonomous driving in school zones.

Kim Sung-han, former national security adviser and Korea University professor emeritus, cited the Israeli city of Haifa as an example. Haifa is known for hosting the research and development centers of Apple, Google and Microsoft, as well as many leading AI hardware startups.

“News reports show that U.S. drone companies are increasingly flocking to Taiwan because it has fewer regulations than Korea,” Kim said. “We are missing out on tremendous opportunities. That's why we need a city like Haifa in Korea, a dedicated AI model city.”

Bone AI CEO Lee Do-kyung also agreed that the lack of regulatory visibility and opportunities to test robots and drones in real-world environments remain as major challenges.

Lim stressed that regulations should not excessively restrict technological development.

“In the 1990s, the U.S. adopted safe harbor policies that protected companies from liability if they took reasonable measures to prevent harm. This reduced uncertainty, attracted investment to the U.S. and helped fuel the internet boom,” Lim said.

“We are already seeing a surge in AI-driven cyberattacks, but we should not simply respond by imposing fines every time an incident occurs … We need to adapt our laws, institutions, policies and the way society thinks to the era of AI transformation.”