This proposal is related to the artificial intelligence (AI) technology, which is showing rapid technological development and having a large social impact, and provides strategic recommendations for basic research to create next-generation AI models that go beyond follow-on development and application development of the current foundation models and generative AI, which are becoming particularly active.

AI technology has developed remarkably, and in particular, generative AI based on the foundation model created by very large-scale deep learning, such as ChatGPT, which appeared at the end of November 2022, has come to exhibit extremely natural dialogue response performance and high versatility and multimodality. This is expected to have a significant social impact as it brings about a rapid transformation of human intellectual work in general and has a broad ripple effect in various fields, including industry, R&D, education, and creative work. According to McKinsey & Company^1）, generative AI could bring trillions of dollars worth of value to the global economy annually. For Japan, where the working population is shrinking, it is greatly expected to increase productivity and bring about industrial and economic growth, and so follow-on development and application development is gaining momentum while U.S. big tech companies are far ahead.

On the other hand, various ELSI (Ethical, Legal, and Social Issues) concerns have been raised, including misuse for faking and spoofing, hallucination (the phenomenon of plausibly returning lies), increased social bias, and copyright infringement. In addition, since it will be a general-purpose fundamental technology used in various activities, excessive reliance on services provided by foreign big tech companies will create risks in terms of economic security and the international competitiveness of scientific research and industry.

In response to the above expectations and concerns, policy formulations related to generative AI were rapidly launched in various countries, both in terms of measures to promote innovation and rulemaking against risks. Social principles related to AI have been discussed and formulated at the national and international level in 2019, including the Japanese government's "Principles of Human-centric AI Society" and the OECD Principles on AI. After that, the phase is shifting from principles to practice, and now rulemaking related to generative AI has become an important issue. This was also discussed during the G7 Hiroshima AI process, which Japan chaired.

As mentioned above, efforts to follow-on development and application development of the current foundation models and generative AI are gaining momentum, and policy efforts to develop rules to address pressing issues are underway. In light of this situation, this proposal focuses on the strategic enhancement of basic research to create the next-generation AI models that go even further.

In developing this strategy, we set the following courses of action by identifying promising directions through a bird's-eye view of related trends, interviews with more than 50 experts, and exchanges of opinions and discussions at workshops and academic conferences.

1. We consider the development toward the next-generation AI models not only in terms of the development of the functionality and performance of the AI itself, but also in terms of the development of the relationship between the AI and others (humans and other AIs) or society. Because, while current AI models have problems with resource efficiency, real-world operation, logic, reliability, and safety, they show phenomenal performance and are changing human activities and social structures. It will be necessary to consider the requirements for the next-generation AI models not only in terms of improvements to current problems, but also in terms of the nature of the relationship between AI and humans or society.
2. We acknowledge the diversity of approaches toward the next generation AI models while encouraging the sharing of awareness and knowledge of the issues among them. This is because exploring a wide range of possibilities for technological breakthroughs while encouraging synergy and fusion, rather than disparate efforts, will accelerate basic research.
3. AI technologies to solve social issues facing Japan and to support social development taking advantage of Japanese characteristics should be owned by Japan. Rather than competing with big tech companies from other countries, it is necessary for Japan to overcome the limitations and problems of the current AI model for the development of Japanese society and people's wellbeing in the face of a declining workforce, while ensuring economic security for our country.
4. We try to design a new research ecosystem to promote basic research, considering that the style of AI research is changing, such as the shift to big science, and that big tech companies are taking a large lead in basic research, and that efforts based on comprehensive knowledge are becoming indispensable.

Based on the above courses of action, the following (1) to (4) are important R&D issues that should be addressed in basic research for next-generation AI.
（1）Research on basic principles and basic architecture of next-generation AI models.
（2） Research on technologies to deal with the risk of incompatibility with humans and society arising from the development of AI models.
（3） Research on scientific research and problem-solving process innovation linked to the development of AI models.
（4）Research on the relationship between AI, humans, and society and how it should be.

While (1) is the core of the next-generation AI model design, it is essential to simultaneously address (2), techniques for dealing with AI risks. While (1) and (2) are basic research that are inseparable, (3) is process innovation research, which is basic research to create social value from (1) and (2). In addition, (1), (2), and (3) are mainly research on information science and technology, while (4) is research in the humanities and social sciences on the state of AI society. While the technical requirements and guidelines for (1), (2) and (3) are established by (4), the concept and goals of (4) can change as a result of the technological development of (1), (2) and (3). These four factors are interrelated and progress in tandem.

In promoting these efforts, it is necessary to take into account the changes in R&D style as mentioned in Course of action D. In other words, R&D in the field of AI, even in basic research, is becoming increasingly big science, high-speed/high-impact, and closed. It is necessary to consider how the R&D system, infrastructure, and programs should be under these circumstances, and to create a research ecosystem that supports them. These include the following elements.

1. Formation of a research ecosystem where various research institutions cooperate with each other
2. Continued operation and enhancement of large-scale computing facilities for shared use.
3. Developing a system for aggregating, sharing, evaluating and managing AI models and multimodal data.
4. Open international collaboration in basic research and rulemaking and the establishment of a support system.
5. Program design to promote the active participation of not only researchers in the technical fields but also in the humanities and social sciences.
6. Establishing an organization to serve as a hub for the research ecosystem.
7. Flexible and agile program management that leverages the research ecosystem.
8. To secure and develop human resources to support the research ecosystem.