The competition for artificial intelligence (AI) hegemony is now expanding into the physical world a..
https://www.mk.co.kr/en/it/12034463
Publish Date: 2026-05-03 03:02:00
Source Domain: www.mk.co.kr
사진 확대
The competition for artificial intelligence (AI) hegemony is now expanding into the physical world and into agents. This means that South Korea, which is seeking to become a third-tier AI player after the U.S. and China, has more technologies that must be secured at the national level. Beyond short-term commercialization, the importance of mid- to long-term R&D is also growing more than ever.
Hong Jin-bae, head of the Information and Communication Planning and Evaluation Institute (IITP), who leads AI and information and communication technology R&D, recently said, “We plan to invest about 1.3 trillion won in R&D this year,” adding, “AI models and semiconductors will be key areas.”
Regarding the recent technological trend, Director Hong said, “Now it is completely changing to intelligent AI that acts in the era of Generative AI. Agent innovation has emerged on the screen and physical AI has emerged in the real world off the screen, he said. “We are focusing on supporting the things we need to develop in this technology flow.”
The AI model is where IITP’s focus is first. If the government is developing an “independent AI foundation model” to secure a source model for Korea, IITP is also promoting research on an agentic AI for autonomous action and a world model that understands the physical world.
“Physical AI is largely composed of three axes: robot, autonomous driving, and autonomous manufacturing,” Director Hong said, explaining the future tasks, saying, “Korea is particularly lacking a little ‘brain’ for robots.” One example is the R&D task of the ‘World Foundation Model for Physical AI Model Learning’, which will be invested about 15 billion won this year.
In particular, IITP aims to secure three general-purpose technologies in the physical AI field: a vision, language, and behavior (VLA) foundation model that processes multimodal inputs, a simulation platform based on actual and synthetic…
