As advancements in artificial intelligence (AI) lead to situations beyond human control, global tech giants are advocating for a cautious approach to AI development. Amidst various government responses from countries including the U.S. and China, experts in South Korea are calling for a 'two-track' strategy that promotes the development of frontier models while ensuring a verified AI ecosystem through safety and security measures.
Lee Won-tae, chair of the Security Special Committee, stated on October 1 during a seminar hosted by the National AI Strategy Committee on the topic of 'AI Development Pace Responsibility,' "As global companies raise concerns about 'pacing,' discussions are underway at the national level, and South Korea must clarify its stance on 'responsibility for pace.'"
The concept of pace responsibility emphasizes the need to build both frontier capabilities and safety and security controls simultaneously.
Last month, Dario Amodei, CEO of Anthropic, mentioned the 'pacing theory' but clarified that it does not imply halting AI development. He noted, "It is not about slowing down AI development but rather ensuring that we do not leave it unchecked to the point of losing control."
Following Amodei's remarks, three major global tech companies launched new models. Lee criticized this dual approach of acknowledging AI risks while simultaneously protecting their markets, asserting, "South Korea must not slow down its AI development." He emphasized that the country should strive to develop frontier models without compromising security.
Lee also observed that the current AI industry is progressing along two axes: one where the U.S. and China are securing their AI industries through minimal communication channels, and another involving multilateral discussions excluding these two nations.
In this context, he argued that South Korea should formulate its AI strategy based on 'functional solidarity.' He stated, "Rather than choosing sides, we must design common denominators that address their needs to lead international issues and propose alternatives."
He added, "We should not join late after standards are established but rather act as co-designers to create norms by establishing bilateral rules."
Choi Dae-sun, head of Soongsil University's AI Safety Research Center, emphasized the need for institutional and technical foundations to build safety systems while developing frontier models.
He pointed out, "We do not know the extent of risks posed by AI, making it difficult to define preparedness levels and scope of response, especially when unintended consequences arise." He stressed the necessity of a response system.
Currently, South Korea is only validating models but needs to establish a system to evaluate overall solutions and revise legislation to control potential misuse.
Kim Sung-hoon, CEO of Upstage, noted that the era is shifting from AI to superintelligence (SI) and is moving toward self-improving superintelligence. He emphasized the need to develop models that can surpass the gap created by domestic pacing in the face of rapid advancements by global tech giants. He stated, "As the U.S. and China race ahead, South Korea must also advance to SI2 and then take responsibility for the pace of development."
To ensure pace responsibility, he suggested that the National Information Society Agency (NIA) should conduct third-party evaluations of performance and limitations, and national safety assessments should be carried out, along with implementing safety measures in self-improvement loops.
Lee Han-wool, vice president of Aim Intelligence, also stressed the need for South Korea's frontier models to emerge while establishing evaluable safety standards. He added that, considering the risks of AI misuse, it is essential to define the scope of model disclosure and access, as well as to review the authority granted to agents and intervention methods in case of issues. He noted that factors complicating the use of verified technologies in security tasks should also be examined.
Lee Won-tae, chair of the Security Special Committee, stated on October 1 during a seminar hosted by the National AI Strategy Committee on the topic of 'AI Development Pace Responsibility,' "As global companies raise concerns about 'pacing,' discussions are underway at the national level, and South Korea must clarify its stance on 'responsibility for pace.'"
The concept of pace responsibility emphasizes the need to build both frontier capabilities and safety and security controls simultaneously.
Last month, Dario Amodei, CEO of Anthropic, mentioned the 'pacing theory' but clarified that it does not imply halting AI development. He noted, "It is not about slowing down AI development but rather ensuring that we do not leave it unchecked to the point of losing control."
Following Amodei's remarks, three major global tech companies launched new models. Lee criticized this dual approach of acknowledging AI risks while simultaneously protecting their markets, asserting, "South Korea must not slow down its AI development." He emphasized that the country should strive to develop frontier models without compromising security.
Lee also observed that the current AI industry is progressing along two axes: one where the U.S. and China are securing their AI industries through minimal communication channels, and another involving multilateral discussions excluding these two nations.
In this context, he argued that South Korea should formulate its AI strategy based on 'functional solidarity.' He stated, "Rather than choosing sides, we must design common denominators that address their needs to lead international issues and propose alternatives."
He added, "We should not join late after standards are established but rather act as co-designers to create norms by establishing bilateral rules."
Choi Dae-sun, head of Soongsil University's AI Safety Research Center, emphasized the need for institutional and technical foundations to build safety systems while developing frontier models.
He pointed out, "We do not know the extent of risks posed by AI, making it difficult to define preparedness levels and scope of response, especially when unintended consequences arise." He stressed the necessity of a response system.
Currently, South Korea is only validating models but needs to establish a system to evaluate overall solutions and revise legislation to control potential misuse.
Kim Sung-hoon, CEO of Upstage, noted that the era is shifting from AI to superintelligence (SI) and is moving toward self-improving superintelligence. He emphasized the need to develop models that can surpass the gap created by domestic pacing in the face of rapid advancements by global tech giants. He stated, "As the U.S. and China race ahead, South Korea must also advance to SI2 and then take responsibility for the pace of development."
To ensure pace responsibility, he suggested that the National Information Society Agency (NIA) should conduct third-party evaluations of performance and limitations, and national safety assessments should be carried out, along with implementing safety measures in self-improvement loops.
Lee Han-wool, vice president of Aim Intelligence, also stressed the need for South Korea's frontier models to emerge while establishing evaluable safety standards. He added that, considering the risks of AI misuse, it is essential to define the scope of model disclosure and access, as well as to review the authority granted to agents and intervention methods in case of issues. He noted that factors complicating the use of verified technologies in security tasks should also be examined.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.