In recent years, the application of artificial intelligence (AI) technology in various real-world fields has been progressing with the evolution of deep learning technologies. Particularly in Japan, a decline in labor productivity due to the declining population has become a societal issue. AI technology is anticipated to be a useful means to resolve such issues. However, in fields such as medical care, transportation, and education, where AI reasoning results have significant impacts on society and the economy, the introduction of AI technology is still limited due to explanatory and reliability issues regarding AI reasoning. The construction and development of AI technologies generally assume large amounts of data. As such, application in cases where gathering large amounts of data is difficult, for example in manufacturing and social infrastructure operations, or where large amounts of data can be collected but annotation processing costs are high, the introduction of AI technology in the real world has proven challenging. Against this background, the NEDO “Technology Development Project for Next-Generation Artificial Intelligence that evolves with people” was launched as a five-year project from the 2020 fiscal year to the 2024 fiscal year (Basic Plan). Research and development activities (R&D items) under this project are as follows.
R&D Item 1: Development of basic technologies for AI systems that evolve with humans
To realize the development of basic technologies for AI that can explain the reasoning process and reasoning rationale, as well as basic technologies for AI that understand and learn human intentions and knowledge.
1- ① Development of a framework for AI systems that evolve with people
To develop a framework for “AI systems that evolve with people” which will be a reference for many fields. Specifically, basic frameworks will be developed, such as subsystems that comprise the “AI system that evolves with people” and the functions that each subsystem should share, the relationships between each subsystem and its functions within the system, and the input/output relationship between the system and the external environment—while simultaneously developing AI systems that actually apply the said framework.
1- ② Development of basic technology for AI which can explain
In AI systems that evolve with humans, it is necessary to be able to determine the cause of AI misjudgment and to make it possible for people to determine the validity of AI decisions. In order to achieve this, basic AI technologies will be developed which can explain the results of AI learning as well as the reasoning process, rationale, and scope, while maintaining the necessary accuracy.
1- ③ Development of basic AI technologies that understand and learn human intentions and knowledge
In an AI system that evolves with people, AI must understand the intentions and knowledge of people and learn according to the data. To achieve this, basic technologies will be developed, such as technologies that automatically structure knowledge from document information, manuals produced by people, and technologies that enable the writing and reproduction of tacit human knowledge.
Research and Development Item 2: Establishing a method of evaluating and managing AI that can be trusted in the real world
Realize items such as evaluative items indicators, specific quality evaluation with identified aims, stipulation of management guidelines and inferred result stability measurement technologies, quality evaluation such as technology improvements, and development of management methods.
Research and Development Item 3: Development of AI that can be easily constructed and introduced
Realization of development of technologies for efficiently building generic trained models with different types of data including images, videos, and languages; development of semi-generic trained models divided according to actual application fields; development of technologies for building high-precision models from small amounts of data in various fields; development of technologies for efficiently managing data and constructed models to facilitate utilization.
NEDO Project Mission for the Development of Next-Generation Artificial Intelligence that evolves with people: While the application of AI technology has already been expanding in various fields and tasks, the application of AI technology has been limited in areas and tasks with significant social and economic impact, such as manufacturing, medical care, and transportation. In order to expand the application of AI technologies in these fields, it is important that people and AI work together in their respective areas of expertise to realize an “AI system that evolves with people” in which people and AI grow and evolve together. This project will develop the basic technologies for “AI systems that evolve with people.” In addition, we will establish methods for evaluating and managing AI systems and develop AI technologies that can be easily constructed and introduced so these technologies can be smoothly implemented in society.
The NEDO enterprise carries out a wide range of research in information infrastructure, infrastructure, natural language processing, audio processing, and image processing for these research and development items. The current research project is positioned within 1- ② “Development of basic AI technologies capable of explanation.”
AI’s explanatory capability regarding educators as well as students is a real issue in education. The reliability of the AI-judgment-based ability evaluation and the satisfaction with educational feedback provided by AI is anticipated to guarantee the quality of education itself.