South Korea Ranks High in AI Metrics, But Credit Belongs Elsewhere
South Korea ranked third globally in notable AI models released last year and first in AI patents per 100,000 people, according to Stanford University's Human-Centered AI Institute. The Ministry of Science and ICT released these findings in late April.
Government officials immediately claimed credit. Deputy Prime Minister Bae Kyung-hoon, who also serves as Minister of Science and ICT, said the rankings reflected "full-scale national support" since the Lee Jae Myung government took office. A Cheong Wa Dae official praised the achievement as a result of government efforts made in just eight months.
Stanford's AI index manager rejected that framing. Sha Sajadieh told an inquirer that the report "does not attribute achievements to specific policy periods or government efforts." She added that South Korea's strength "reflects activities accumulated over at least several years across the entire ecosystem."
The R&D Timeline Problem
AI models typically take one to two years to develop. Foundational research precedes that work. Creating a usable AI model in eight months is not feasible with current technology.
The patents measured in this year's report were granted through 2024, with underlying filings dating back years before that. The notable models reflect similarly long development cycles. Neither metric can be attributed to a government that took office recently.
Where the Work Actually Happens
Industry produced over 80% of notable AI models globally in 2025. South Korea follows this pattern, with its notable models and top AI talent concentrated in the private sector, particularly in hardware and semiconductor specialization.
Government can shape the environment for private research through compute resources, talent development, and regulation. But isolating the effect of specific policies on model or patent output falls outside what Stanford measures.
Deputy Prime Minister Bae worked at LG AI Research before government service. A Cheong Wa Dae official came from Naver. Both understand that multi-year research pipelines precede public results. Yet both attributed success to government action.
The developers who built these models and filed these patents worked on them for years. That work deserves the credit.
Learn more: AI for Government and AI Research resources can help government professionals understand how AI development actually works.
Your membership also unlocks:
