Building an AI research pipeline: MEXT leading scientific revival — ARiSE and SPReAD at the core

The Ministry of Education, Culture, Sports, Science and Technology (MEXT) will include the construction of a next-generation research infrastructure supporting "AI for Science" in its budget blueprint for the upcoming fiscal year.

Various nations are envisioning and planning next-generation infrastructures through the introduction of AI agent groups, such as the American Science Cloud in the United States, the KRDC Ecosystem in South Korea, and the EOSC in Europe. Japan aims to realize a scientific revival by swiftly constructing its own pipeline for AI for Science.

In the basic strategic policy for promoting AI for Science compiled by MEXT last year, the five-year period starting in 2026 has been designated as an intensive reform period. By setting 20 specific actions and accelerating efforts with a sense of speed through bold investments, Japan aims to establish its status as an advanced nation in AI for Science.

Target examples include raising the number of AI-related papers among the top 10% of highly cited papers to 3rd in the world, increasing the number of advanced AI research personnel by 3,000 within five years, expanding the capacity of the NII Research Data Cloud (NII-RDC) fivefold by fiscal year 2030, doubling the speed of SINET by fiscal year 2028, and increasing shared computing resources for AI for Science tenfold or more by fiscal year 2030.

One of the initiatives for this purpose is the Innovation Program for Scientific Research via AI for Science. This consists of the ARiSE initiative (1 to 3 billion yen per project for 3 years), which fosters the top tier of talent and nurtures research fields that leverage Japan's strengths, and the SPReAD initiative (5 million yen per case for 1,000 cases), which enables researchers in all fields to utilize AI to advance and accelerate their research. ARiSE began accepting public applications on the 12th of this month, and SPReAD is scheduled to launch its second call for proposals in early June. Additionally, regarding the handling of research data related to such AI utilization, a checklist is being created to determine what should and should not be disclosed.

For instance, it verifies whether data is structured to prevent unauthorized use outside its intended purpose, confirms the locations where research data is handled, and ensures appropriate measures are taken given the possibility that training data could be inferred or reconstructed from AI models. Because these practices have the potential to expand into research across multiple disciplines, cross-sectional expansion is also being considered as the project progresses.

Meanwhile, the AI White Paper published by the Liberal Democratic Party clearly positions AI for Science as a cross-cutting priority item in the AI Basic Plan and the Integrated Innovation Strategy 2026. Centering on priority areas, it states that agile and bold investments of a scale and speed matching global standards will be made over multiple years (1 trillion yen over 5 years) for the development of AI agents and AI-driven research systems.

Efforts toward AI for Science are also intensifying internationally.

The United States has launched the Genesis Mission, which aims to fundamentally reform scientific research and technological innovation through AI. To double the productivity and impact of US technological innovation over 10 years, the Department of Energy (DOE) announced an initial investment exceeding 320 million dollars.

The UK's AI for Science Strategy has set its first mission to utilize AI to "create trial-ready drug candidates within 100 days" by 2030, investing approximately 137 million pounds.

Taking these circumstances into account, MEXT intends to request a substantial budget increase for AI for Science-related expenses.

While strategically reinforcing experimental foundations (such as forming 3 or more shared automated experimental hubs), data foundations (such as a fivefold or greater enhancement of shared storage), and computing resources (such as a tenfold or greater enhancement of shared computing resources), MEXT will build an all-Japan next-generation research infrastructure by developing and organizing a system (pipeline) that connects these foundations in a high-speed, highly reliable, and seamless manner.

The construction of foundations such as a common authentication system, a secure and high-speed communication network, and highly transparent and reliable AI agents will be included in the budget request. Furthermore, as the number of applications for SPReAD is extremely high, the budget will be expanded to allow for the selection of more research projects. In addition, the construction of a mid-scale research program bridging SPReAD and ARiSE is also under consideration.

Clear basic rules vital for researchers gaining greater autonomy

In a survey conducted by Elsevier (covering over 3,200 researchers across 113 countries), only 27% felt they had received sufficient training regarding AI utilization, and only 32% answered that their institution's AI governance was good. The results were particularly low among Japanese researchers, with training at 20% and governance at 26%.

Utilizing AI in research secures more research time and enables researchers to demonstrate their creativity. However, leaving unclear basic rules, such as the permitted applications of AI and their boundaries, could lead to information leaks or violations of journal regulations. It will be necessary for universities, research institutions, the government, and academic societies to work together to address this.