'US scores 100 on AI, China 70, Korea 35' — but physical AI is a different game, deputy minister says

"If we set the United States as 100 on AI, China would be around 70 — and Korea around 35."

Ryu Je-myung, second vice minister of science and ICT, put South Korea's generative AI capability at roughly one-third of the United States' level. But he argued the picture looks very different for physical AI, where no clear winner has yet emerged. With world-class manufacturing foundations in semiconductors, automobiles and robotics, Korea can realistically aim to become the global leader in that field, he said.

The government plans to develop a general-purpose robot foundation model — the "brain" of a robot — by next year, and to advance it to a level capable of handling highly complex tasks by 2029.

Ryu made the remarks Thursday at the 290th Korea Enterprises Federation Forum held at Joseon Hotel in Seoul, where he delivered a lecture titled "National Future Industries and Physical AI." He said the government had set a goal of becoming the world's top power in physical AI, noting that while Korea entered the broader AI race late, competition in physical AI is still in its early stages.

Physical AI refers to technology that combines AI — which previously generated and reasoned over information on screens — with physical bodies such as robots, self-driving cars and drones, enabling machines to perceive and act in the real world. The government is pursuing physical AI as one of three "mega projects," alongside semiconductors and AI data centers, aimed at transforming domestic industry.

Ryu began by acknowledging the reality Korea faces in generative AI. "Korea's pace of development in generative AI is about one-third that of the United States," he said. "China is right on America's heels, and we have not yet reached that level."

He said Korea nonetheless has an opportunity in physical AI. While conventional generative AI learned from text on the internet and accumulated human knowledge, physical AI must master the physical laws of the real world — temperature, humidity, gradient, wind — as well as the vast data generated during actual work processes. Factories and production lines themselves, he said, become enormous training grounds for AI.

"For physical AI to enter manufacturing sites, it must learn on those sites — and no country outside China has a manufacturing base like ours," Ryu said. He added that Korea's strong manufacturing foundation, industrial infrastructure and irreplaceable semiconductor production capacity, taken together, make the goal of global leadership in physical AI achievable.

The government is also accelerating development of a robot "brain." The Ministry of Science and ICT will develop a general-purpose robot foundation model, while the Ministry of Trade, Industry and Energy will build specialized models that can be applied directly to logistics, manufacturing and other real-world settings — a two-track approach.

"The Ministry of Science and ICT's goal is to build a model capable of handling highly complex tasks by 2029," Ryu said. "By next year, we aim to produce a model on par with the foundation model Nvidia is currently developing."

With robots now operating in physical spaces, the government views domesticating the robot "brain" as directly tied to industrial security. Once physical AI is deployed in earnest on factory floors, robots will see the inside of facilities firsthand, learning companies' production know-how and accumulating sensitive data.

Ryu cited the example of Chinese-made self-driving cars transmitting data collected during operation to servers in China. "Imagine placing a Chinese robot on a factory floor and having it send all the know-how it observes back to China — wouldn't that be quite a nightmare?" he said.

He added that Korea must not depend on China or any other country even on the hardware side, and said the case for self-reliance is even stronger when it comes to the brain. "You cannot hand your own mind over to someone else," he said.

Hardware competitiveness, however, remains a challenge. Korea deploys industrial robots at one of the highest rates in the world, but the domestic content rate for key components is low, leaving the country heavily dependent on Japanese and German suppliers, Ryu said. The government plans to pursue localization of core components — including actuators, robotic hands and sensors — while also working to secure the data needed to train physical AI systems.

Ryu also pointed to the standing Korean companies have already built in the global AI industry as a foundation for competing in physical AI.

He recalled helping organize an AI summit held in San Francisco last month, where Nvidia CEO Jensen Huang, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei gathered in one place. "Each of these individuals is extraordinarily difficult to bring together," Ryu said. "That scene showed just how powerful the capabilities our companies have built have become." He said Korea's competitiveness in semiconductors, automobiles and manufacturing is the core asset drawing global AI companies into partnership with Korea.

A question-and-answer session after the lecture touched on competition with China and the country's 52-hour workweek cap.

When an attendee asked whether Korea could compete given that Chinese companies pour resources into research and development under the so-called "996" model — working from 9 a.m. to 9 p.m., six days a week — Ryu drew a line at directly evaluating the policy.

He described a recent visit to Huawei's research and development center in Chengdu, China. "Looking at the scale of investment, the caliber of the people, and the hours and environment they are putting in, I thought Korea needs to do a great deal of soul-searching as a society," he said.

He said domestic AI companies are working hard and finding their own solutions within the constraints of Korea's legal and social environment. "I hope society arrives at good answers to these fundamental questions, so that we have an environment where we are not left behind in this competition," he added.

On whether the 52-hour workweek limit should be eased, however, Ryu declined to take a position. "I don't think it would be appropriate for me to comment on that area," he said.