*Editor’s note: K-VIBE invites experts from various K-culture sectors to share their extraordinary discovery about the Korean culture. 

 

Matthew Lim's AI Innovation Story: Developmental Golden Time for Korean-Style AI

 

By Matthew Lim, AI expert and director of the Korean Association of AI Management (Former Head of Digital Strategy Research at Shinhan DS)

 

 

 

The acceleration of the global artificial intelligence (AI) technology competition is in full swing. However, the position of South Korea is gradually shrinking.

 

At the beginning of the development of generative AI, South Korea aimed for a spot in the top three globally, but it has now slipped to a mid-range position.

 

There is a global AI index (GAI) evaluated by the British media company Tortoise, which has assessed the AI capabilities of 83 countries worldwide since 2019. The GAI is divided into three main categories—implementation, innovation, and investment—and further subdivided into seven items (talent, infrastructure, operational environment, research, development, government strategy, commercial ecosystem), and 23 sub-items, making a total of 122 indicators. In the 2024 GAI index, South Korea ranked 6th out of 62 countries. The gap between the number one ranked United States and the second-ranked China is over 46 points, with the United States holding a dominant lead.

 

In the evaluation by Stanford University’s Human-Centered AI Institute (HAI), South Korea ranked 7th out of 36 countries.

 

▲ This image of "Artificial Intelligence Index (AI Index) 2024" is captured from Stanford University's website. (PHOTO NOT FOR SALE) (Yonhap)

 

In the AI maturity matrix published by the Boston Consulting Group (BCG), the performance is even less impressive. South Korea, along with Germany and Japan, remains in the “Steady Contenders” group, a second-tier category. In the current era where AI technology has become a core element of national competitiveness, this is a serious warning sign.

 

Securing a Korean-style large AI model has been like a “dream unattainable” for South Korea. The lack of computing infrastructure, including GPUs, and difficulties in securing advanced AI talent have continued to hinder progress. Despite possessing world-class semiconductor technology, South Korea has ironically lacked the essential infrastructure for the AI era, including AI semiconductors and computing power.

 

In this context, the “DeepSeek R1” model, released by Chinese company DeepSeek, offers a new breakthrough for South Korea. This model achieved a 79.8% pass rate in solving math problems, meeting the qualification for the American Math Competitions (AMC) preliminary round, AIME.

 

In the “MATH-500,” which evaluates performance on 500 math problems, the model achieved an astounding 97.3%, a performance comparable to OpenAI’s latest model, “ChatGPT o1,” which scored 96.3%.

 

What is even more noteworthy is that this model achieved such performance with relatively fewer computing resources. The concerns raised recently about privacy protection are primarily issues that could arise during the service delivery process. In other words, these concerns should be viewed separately from the technical achievements of the model itself.

 

The true significance of the birth of DeepSeek R1 lies in the fact that this model has been made open-source.

 

▲ This Yonhap file photo shows DeepSeek-generated info. (Yonhap)

 

The urgent need for the development of a Korean-style AI model is clear.

 

First, we must secure differentiated competitiveness in the global market. In a situation where AI is leading innovation across industries, without independent AI technology, South Korea risks losing its competitive edge in future markets.

 

Second, we must achieve an efficient digital transformation across domestic industries. Only an AI model that deeply understands the characteristics and culture of South Korea's industries can drive the successful digital innovation of Korean companies.

 

Fortunately, we still have a powerful weapon: world-class data assets. These include medical data accumulated through the national health insurance system, manufacturing data collected from the semiconductor and automobile industries, and transaction data generated by the world’s top digital financial infrastructure.

 

This high-quality data is a core asset that determines the performance of AI models.

 

Recently, the government has recognized the seriousness of this situation and is taking swift action. With the goal of entering the top three global AI powerhouse rankings by 2027, the government is promoting large-scale investments in AI semiconductors and supercomputing infrastructure.

 

A particularly noteworthy policy is the focus on cultivating advanced AI talent. Various approaches are being made, including expanding AI-specialized graduate schools, improving visa systems to attract top overseas talent, and strengthening industry-academia-research collaborations to foster practical talent.

 

In this situation, the open-source release of DeepSeek R1 has presented a new opportunity for South Korea’s leap forward. This model, which achieves high performance with relatively fewer computing resources, can already be fully utilized with South Korea's current infrastructure.

 

Furthermore, by training this model on South Korea’s unique specialized data, we can create a Korean-style AI model. This could be a stepping stone for South Korea to take another leap in the global AI competition.

 

Of course, organic collaboration between the government, companies, and research institutions is essential. AI research support at the national level, relaxation of data regulations, establishment of data-sharing platforms by companies, and talent cultivation through industry-academia-research collaborations must all be systematically implemented.

 

In particular, it is important to create an ecosystem where even small and medium-sized enterprises (SMEs) can effectively utilize AI.

 

With the ongoing global AI market leadership competition, there is no time left to hesitate. We must face the reality of South Korea’s current mid-tier position, maximize our strengths, and create an opportunity for a counterattack.

 

We must not miss the potential shown by DeepSeek R1.

 

Securing AI sovereignty is not optional but essential. To protect the future competitiveness of our industries, we must act now to control our data and technology.

 

The path South Korea must take is clear. Based on efficient open-source AI models like DeepSeek R1, we must create a Korean-style AI model specialized with data.

 

To do so, we need to establish an integrated platform for industry-specific data to efficiently collect and utilize high-quality data.

 

Then, a “two-track” strategy is needed: relaxing regulations on AI model development and usage while strengthening data security. Finally, a national support system must be built to help SMEs easily adopt and utilize AI.

 

Securing world-class AI capabilities is no longer a matter of choice. It is a crucial task for the survival of South Korea’s industries. The emergence of efficient open-source models like DeepSeek R1 has opened a new window of opportunity for South Korea.

 

Ultimately, we must secure our own large-scale AI model, but for now, we must take steps forward through these realistic alternatives.

 

Now is the time to start.

[ⓒ K-VIBE. 무단전재-재배포 금지]