[The National AI Caught in the Originality Trap] Originality vs. Performance...The Government Faces a Dilemma ③

Clarify Evaluation Criteria, Ensure Transparency in Verification
The Project Should Not End as a One-Off Survival Contest
Policy Pathways Needed for Continued Research and Challenges

"If we prioritize uniqueness above all else, performance issues may arise. Conversely, if we focus on performance, we are forced to accept the dilemma of not meeting the uniqueness criteria." (Participant in the independent artificial intelligence (AI) foundation model project)

As the Ministry of Science and ICT announced plans to select two consortiums for the “Dokpamo” project by the end of this year, the AI industry is grappling with the standards for uniqueness. There are conflicting opinions: one side argues that "since the government is investing massive resources such as graphics processing units (GPUs), it is essential to have our own unique architecture (blueprint)," while the other side contends, "Developing a unique architecture is not impossible, but making uniqueness an absolute standard ignores the rapidly changing realities of large language model (LLM) development."

Previously, during the first round of the “Dokpamo” project selection process for five elite teams, controversy over uniqueness led the government to change its original plan of eliminating one team, ultimately deciding not to select two teams instead. The government later clarified, "Open-source utilization is permitted in the first round of evaluation, but teams must prove their own technological capabilities." Despite this, debates among LLM developers in the IT industry remain unresolved.

Given Korea’s urgent need to catch up with the United States and China, some experts advise that a careful approach is needed-one that considers both uniqueness and performance. Wi Jeonghyeon, Dean of the College of Virtual Convergence at Chung-Ang University, pointed out, "Even global models actively utilize open source when necessary. If we develop models from scratch on a zero-base, we risk falling behind."

Sudden changes to evaluation criteria or competition rules have also sparked controversy over fairness. An AI industry executive, who requested anonymity, stated, "Going forward, the government must actively reflect industry opinions to correct the issue of abruptly changing competition rules and standards." The executive added, "With many domestic LLM experts already participating in the Dokpamo consortium, it is also important to ensure that technical verification is conducted objectively and without bias." The government must clearly define its priorities-technological capability, symbolic value, or industrial impact-so that the Dokpamo project can lay the foundation for fostering Korea’s AI ecosystem.

Although the Dokpamo project has attracted significant public attention by adopting a survival-based format, some argue that the focus should not be solely on survival or elimination. Instead, it is more important to create an environment where startups can continuously challenge global LLMs. Shin Jaemin, CEO of Trillion Labs, emphasized, "Startups inevitably face relative shortages of capital and manpower, but even in such circumstances, we are striving to develop unique designs." He added, "We are improving development efficiency, and I believe such challenges will ultimately become the driving force for the advancement of the AI ecosystem."

Uniqueness Controversy Overheating... Fostering the Domestic AI Ecosystem Is Crucial

The government's internal deliberations ahead of the second round of selections are even more complex. Beyond short-term performance competition, the core challenge is how to nurture the domestic AI ecosystem and ensure the practical use of its outcomes through this process.

Kim Kyungman, Director General of the Artificial Intelligence Policy Bureau at the Ministry of Science and ICT, stated, "The purpose of creating an independent foundation model is not simply to produce a single result." He added, "It is more important to establish a structure where various companies and researchers participate in the development process, and the results are later utilized in actual industries and the public sector." This is why the government believes the project should not end as a one-off survival competition.

In particular, how to support companies that fail to pass the evaluation is a key issue the government must address. Since this project has become symbolic of a national AI initiative, there are concerns that a structure where selection or rejection is immediately equated with success or failure could discourage future challenges.

A senior official at the Ministry of Science and ICT explained, "Even companies that are not selected have built up technological capabilities and experience, so there is an internal consensus that policy measures are needed to allow them to continue with follow-up research and challenges." The official added, "We are considering various options, including a form of 'participation award' incentive." Such incentives may include opportunities to utilize GPUs, connections to follow-up research projects, and support for open-source utilization.

Image generated by Gemini

The confusion surrounding the uniqueness criteria is another issue the government must address. The government views uniqueness not simply as "excluding foreign products," but as a matter of sovereignty and control. As AI becomes a universal technology, it is necessary to have models that we can control in areas such as national defense and security. It is also essential that we are responsible for how these models are developed and managed, while simultaneously improving their performance.

A senior official at the Ministry of Science and ICT commented, "Uniqueness has been the minimum requirement for this project from the beginning, but it is true that the industry’s understanding has varied." The official added, "During the recent announcement of additional elite teams, the government shared its interpretation of uniqueness from technical, policy, and ethical perspectives with the industry, and after hearing their opinions, many responded that 'this is sufficiently clear.'"