Launch of Next-Generation AI+S&T Core Technology Development Project to Tackle New AI Architecture Design

[Financial News] The government will secure core technologies that enable artificial intelligence (AI) to use not only data but also physical and mathematical principles to interpret and predict scientific phenomena more accurately and reliably. The goal is to develop AI beyond a simple research assistant into a core technology that accelerates scientific discovery and research innovation.
The Ministry of Science and ICT announced on the 24th that it held a kickoff briefing for the "Next-Generation AI+Science and Technology (S&T) Core Technology Development Project" and has begun full-scale research and development on four newly selected tasks.
AI has recently been used rapidly in scientific research and industrial settings, but most current AI systems work by learning from vast amounts of data and predicting outcomes. As a result, in scientific and engineering fields where physical laws and mathematical principles are crucial, it has been difficult for AI to explain why it produces certain results. It has also faced limits in accuracy under new conditions that were not included in training.
The ministry also plans to challenge the design of new AI architectures that go beyond the limits of existing AI models. The research projects are divided into two areas: physics- and mathematics-based AI interpretation and prediction models, and next-generation AI architecture research. A total of four projects, two in each area, have been selected. The project will receive 20 billion won this year and a total of 20 billion won from 2026 to 2031. For the selected projects, computing infrastructure such as graphics processing units (GPU) will be provided, and the research data and AI models developed through the project will be opened through a public platform.
Specifically, in the physics- and mathematics-based AI interpretation and prediction model area, two projects will be carried out to develop AI models that remain stable even when conditions change. A team led by Woo-seok Ha at KAIST will develop a mathematics- and physics-based causal AI model that infers the causal structure and governing equations embedded in data, even when conditions and environments differ.
A team led by Yu, Jaesok at DGIST will develop an AI Mechanician that visualizes which physical laws are dominant in various phenomena, allowing AI to judge for itself and choose an interpretation strategy. Through this research, AI is expected to produce consistent results that align with physical laws even when conditions change. That could reduce the cost and time required for design, analysis and verification, without having to start validation from scratch each time conditions shift.
In the next-generation AI architecture research area, two projects will focus on newly identifying and designing the learning and structural principles of AI models. A team led by Min-hwan Oh at Seoul National University will mathematically identify the structural limitation in current AI models, where computational demands rise sharply as context length increases. The team will then develop a new AI architecture and learning method that can overcome this limitation and process long contexts efficiently.
A team led by Gyeong-suk Yoon at KAIST will mathematically explain the principles by which generative AI models improve performance as they learn. It will also establish a next-generation AI development methodology that can be applied across the entire process, from architecture design to training and reliability verification. This is expected to help design AI models that operate stably while reducing dependence on massive data and computing resources, thereby contributing to greater self-reliance and competitiveness in Korea's AI technology.
Gyeong-suk Yoon, director of basic and foundational research policy at the Ministry of Science and ICT, said, "If we secure AI models based on the laws of physics and mathematics, the rigor and accuracy of scientific research will improve, which will provide practical benefits not only in research settings but also in industries such as semiconductors and batteries." She added, "In addition, securing new AI model architectures will also contribute to the advancement of AI technology itself."
[email protected] Yun Ji-an Reporter