The U.S. led with 109 such models, followed by China with 20 and the U.K. with eight. Even the United Arab Emirates produced 4 foundational models. (Image courtesy of Pixabay/CCL)
SEOUL, April 17 (Korea Bizwire) – While South Korea demonstrates robust human capital for artificial intelligence (AI) development, the country failed to produce any foundational AI models last year and is experiencing an outflow of talent to other nations, according to a new report from Stanford University’s Human-Centered Artificial Intelligence Institute (HAI).
The “AI Index 2024″ report, released on April 15, shows that South Korea held the highest number of AI patents per 100,000 population in 2022 at 10.26, significantly outpacing the United States (4.23) in third place and Japan (2.53) in fourth. Only Luxembourg (8.73) came close.
South Korea also ranked third globally in AI talent concentration at 0.79%, behind only Israel (1.13%) and Singapore (0.88%), based on data from the professional networking site LinkedIn.
However, the country did not develop any of the foundational models that underpin generative AI technologies in 2022. The U.S. led with 109 such models, followed by China with 20 and the U.K. with eight. Even the United Arab Emirates produced 4 foundational models.
Of the 108 AI models designated as “notable” by HAI, 61 originated from the U.S., 15 from China, and eight from France, while South Korea had none on the list. The UAE and Egypt contributed three and two models, respectively.
Notably, South Korea recorded a net outflow of AI talent at -0.3 per 10,000 LinkedIn members, indicating more professionals leaving the country than arriving. This metric had been positive at 0.3 in 2020 before declining in 2021 and 2022, reaching a significant deficit last year.
Luxembourg (3.67) and the UAE (1.48) exhibited the highest inflows of AI talent, while India (-0.76) and Israel (-0.57) joined South Korea in experiencing sizeable brain drains. The U.S. posted a modest 0.40 net inflow.
In terms of private investment in AI, the U.S. dominated with $67.2 billion, followed by China at $7.26 billion. South Korea ranked 9th among countries analyzed with $1.39 billion.
Last year, Google stood out by releasing 18 foundational models, including Gemini, more than any other company. Meta unveiled 11 models, Microsoft nine, and OpenAI seven.
Training these large language models demanded immense computational costs. Google’s latest AI model, Gemini Ultra, required an estimated $191.4 million, while OpenAI’s GPT-4 cost around $78.35 million.
In benchmarks testing AGI (Artificial General Intelligence) capabilities across science, arts, and engineering, Google’s Gemini Ultra and OpenAI’s GPT-4 achieved 59.4% and 56.8% accuracy respectively, falling short of the 82.60% average for human experts.
Kevin Lee (kevinlee@koreabizwire.com)
