[Insight & Opinion] Establishing 'AI Sovereignty' Should Take Priority Over Developing Korean-Style AI

Overwhelming Investment and Talent in the U.S. and China
Korea Must Build Distinction with 'Sovereign AI'
Balancing Data Access and Regulatory Easing

Measures to promote the AI industry have been announced. Support will be provided to secure 18,000 graphics processing units (GPUs) for the National AI Computing Center, and research funding will be offered to attract global key talents. The funds allocated for talent development and securing foundational technologies amount to about 1 trillion won. The goal is to achieve a top-three global position in AI through the development and industrialization of world-class large language models (LLMs). It is a bold and excellent plan. But is it really feasible?

Unfortunately, the realization of the goal of becoming an 'AI advanced country' is not guaranteed. This is not just about the current level of development. Both the scale of investment and human resources are difficult to compare. The investment scale of the 'Stargate Project' announced by the United States is about 730 trillion won. The investment scale for the National AI Computing Center planned by our government is 2 trillion won. Even in terms of GPUs, Meta alone already owns 350,000 units in the U.S.

While the U.S. is overwhelming in investment scale, we cannot match China in human resources. Since 2017, China has established AI undergraduate majors in 535 universities and set up 43 AI specialized schools and research institutes. As of 2024, the number of students enrolled in AI-related departments in Chinese universities is about 40,000. The number of AI researchers in China exceeds 400,000, whereas we have 20,000.

AI competitiveness is determined more by human resources than by the number of GPUs. Although there is a plan to form a 'national AI team' for LLM development, it is meaningless if there are not enough players to begin with. The fact that domestic companies are focusing more on strengthening cooperation with large overseas IT companies rather than building an independent ecosystem is likely a choice made considering the reality.

Of course, we cannot just stand by. This is an AI war taking place on a global scale. Building AI is not merely a matter of technological competition. It is also critical from the perspective of national security. It is natural that there are calls for a national all-out effort. However, the AI hegemony war is rapidly shifting into a competition mainly between the U.S. and China. It is virtually impossible for us to take the lead in the AI industry. A spirit of challenge is necessary, but the goals must also align with reality. Although the world's first Artificial Intelligence Digital Textbook (AIDT) is being introduced in schools starting this year, it is actually hard to find AI digital textbooks in classrooms. This is because there is insufficient budget to properly provide student devices. For various reasons, the adoption rate of AI digital textbooks nationwide is only 32.4%. Grand slogans and goals are at odds with the disappointing reality.

Practically, the priority is to build 'sovereign AI' capabilities. This refers to AI that utilizes our data and infrastructure to understand and reflect our culture, values, social, and political environment. While LLM development is challenging, developing specialized AI in selected specific areas that no one can surpass is something we can achieve.

We must first establish the direction of data protection policies, including personal information. Securing and smoothly utilizing data, often called the 'oil of AI,' is essential for strengthening AI capabilities. However, in Korea, there are still significant restrictions on data acquisition due to regulations such as the Personal Information Protection Act and copyright law. Balancing personal information protection and regulatory relaxation is not as simple as it sounds.

The so-called 'AI Basic Act,' which will be enforced from January next year, mandates that the Minister of Science and ICT establish and implement a basic AI plan every three years to promote AI technology and industry and strengthen national competitiveness. However, in reality, what the government can do is mostly limited to reducing regulations, providing tax benefits, and helping universities nurture more talent. Even these are not easy tasks.

Kim Sang-cheol, Economic Commentator