With little sign of closing the gap with the most advanced artificial intelligence (AI) models from global big tech companies such as OpenAI, Google, and Anthropic, Professor Yoon-Hyun Kim, a world-renowned natural language processing scholar at the Massachusetts Institute of Technology (MIT) School of Electrical Engineering and Computer Science (EECS), emphasized that “South Korea should never stop developing its own AI technology.”
in a recent interview Mainichi Keizai ShimbunProfessor Kim said, “It’s true that it will be extremely difficult to catch up with the highest-performance models of frontier AI in the United States, but that doesn’t mean we won’t try.” He further added, “The very act of taking on the challenge of developing frontier-level large-scale language models (LLMs) can take Korea’s AI technology and talent pool to the next level.”
Discussions around “sovereign AI” (independent AI ecosystems built on a country’s own data and infrastructure without relying on foreign technology) first emerged with the launch of OpenAI’s ChatGPT in 2022. The concept gained renewed attention this year with the release of Anthropic’s ultra-high-performance Mythos model and the US government’s move to restrict the export of AI models.
Professor Kim pointed out that “As AI models become more sophisticated, they will become directly linked to national defense and security issues,” and added that South Korea must “prepare for a scenario in which the U.S. government restricts foreign access to frontier models.” He reiterated that “This is a field in which we must keep trying even if we fail.”
He argued that securing proprietary technology is especially important in an AI agent environment. “Not all agents require front-line level performance,” he explained. “The key is to have a unique model tailored to each agent.” In an age of agents where speed and efficiency are paramount, the technical ability to design small, optimized models for specific services becomes more important than relying on a single large model.
Professor Kim has been collaborating with Naver since 2022 on research to advance AI search technology. Through Naver’s U.S. AI center, Naver Search US, the ultimate goal is to develop an agent that not only optimizes search results but also seamlessly connects Naver’s various services such as shopping and maps. He said, “We are conducting preliminary research to realize “NAVER is all you need,” where everything a user wants can be solved within NAVER.”
Professor Kim is particularly interested in a “health pill” that he aims to launch by the end of this year. “In the health field, even a single hallucination can lead to serious problems, so we’re thinking deeply about how to design the validation layer,” he said.
The paper “FlowBot” co-authored by Professor Kim and Naver Search US and accepted at the 2026 International Conference on Machine Learning (ICML) is one of the results aimed at solving problems in agent design. FlowBot is an engineering technology in which AI autonomously discovers and learns the optimal task sequence when multiple agents work together, eliminating the need for manual workflow design by humans. “Naver has various agents such as search, shopping, etc., and manually creating workflows for each domain is very costly,” Professor Kim explained. “FlowBot is a concept that allows an agent to learn its own workflow, similar to machine learning.”
Regarding Naver’s differentiation, he said, “Google’s main focus is search, but Naver is a portal that covers a variety of areas such as shopping and locations, so the range of problems it can solve within its platform is much wider.”
As a researcher, Professor Kim also has an eye on the next evolution beyond LLM’s current universal structure, the Transformer architecture. “For 10 years, AI has been advancing based on Transformer, but I don’t believe that’s the end of architecture,” he said. “From now on, research into next-generation architecture will be my main focus.”
Regarding the direction of education in the AI era, Professor Kim advised, “Basic abilities such as structural thinking will become even more important.” He emphasized, “To use AI effectively, the ability to formulate your own argument is essential.Education needs to help people understand how AI works, and help them distinguish between tasks that should be left to AI and areas that require human judgment.”
