Bae Kyung-hoon, vice prime minister and Minister of the Ministry of Science and ICT, stressed setting concrete goals and priorities for artificial intelligence (AI) development strategies at Government-funded research institute(s) in the science and technology sector, saying, "If you pursue several projects at the same time, you could end up finishing after doing only verification."
On Jan. 12 at the National Research Council of Science & Technology (NST) in Sejong, Bae said at a briefing on grants research institutes' work, "A large language model (LLM) cannot solve every problem, and multiple technologies are needed," adding, "We first need to set refined goals for where and how to use AI."
Regarding the Korea Institute of Energy Research (KIER)'s proposal for an "AI to improve energy efficiency by 15% based on a language module," Bae noted, "To get agents to reach a meaningful level in each field, a larger-scale approach is needed," and pointed out, "With the current approach, it will be difficult to realize a true agent AI or to apply it on site." He also criticized conversational AI or autonomous facility operation AI in the sense that "even with narrower goals the difficulty is high, and if pushed simultaneously on multiple fronts, it will be hard to deliver results."
On ETRI's report about developing a robot AI foundation model, he added, "Since robot intelligence as well as multimodality and action intelligence are required, there needs to be a clear definition of what you intend to build," and "If you decide clearly whether to move first into immediately feasible areas by forecasting demand, to strengthen decision-making intelligence, or to specialize certain functions, the plan will become more solid."
On ETRI's mention of a general artificial intelligence (AGI) concept, he asked in return, "If you are talking about future AGI, shouldn't the target level be higher?" and suggested a shift in direction in the sense that "if it is not at that level, it is better to leave it to institutions that can do it well and focus on areas where tangible results are possible, such as a physical AI foundation model or specialized models."
Bae cited "data systems" as the first task to tackle in the AI transition of grants research institutes. He said, "Systematically collecting data that AI can read and learn from is the starting point," and suggested that ETRI consider building an "AI data generation platform" that extends to data generation, tuning, and training utilization with expert validation. He continued, "Rather than clinging only to developing foundation models, you need to establish a system that generates data. Even in the United States' Genesis Mission, that part is emphasized as the core," adding, "NST also needs to form a task force (TF) for building AI data at grants research institutes."
From the field came the point that "national research institutes need unique differentiation." Bae also mentioned a comment from the YouTube live chat that suggested, "It needs to be different from private companies' AI, but right now it looks like a hippopotamus in a swamp devouring the expanded AI budget," saying, "We should take that to heart."