Trusted AI

1. Overview

In recent years, AI technologies have been permeating society in various forms, including systems and services using AI technologies. Based on the “Social Principles of Human-Centric AI” (decision made by the Integrated Innovation Strategy Promotion Council in March 2019), the “AI Strategy 2019” (decision made by the Integrated Innovation Strategy Promotion Council in June 2019) which emphasizes on the importance of the development of the “Trusted Quality AI” was formulated in order to promote the social implementation of AI.
On the other hand, deep learning, which is the center of current AI technologies, requires a large amount of supervised training data, and issues related to its reliability and safety have been pointed out, such as improving the explainability and convincibility of results, ensuring fairness, and overcoming vulnerabilities in unknown cases. In addition, there are problems related to reliability of data itself, such as the distribution of fake information and data falsification.
To address these issues, solutions for current AI technologies have become an urgent issue, and the industry has begun to consider solutions by consortiums, etc. It is necessary to develop and innovate AI technologies to overcome their limitations and to secure fundamental reliability that can meet the demands of society.
This strategic objective aims at developing basic technologies for the future evolution and reliability of AI.

2. Targets

This strategic objective is to promote research and development aimed at creating “Trusted Quality AI” based on “Social Principles of Human-Centric AI”. In particular, the following three targets are to be achieved.
(1) Creation of new technologies that overcome the limitations of current AI technologies
(2) Creation of technologies to ensure the reliability and safety of AI systems
(3) Creation of technologies to ensure data reliability and to support decision making and consensus building


Office for Basic Research Projects, Basic Research Promotion Division, Research Promotion Bureau