Japan’s AI Utilization Guidelines: an initiative for implementing the OECD AI Principles

At the first OECD meeting on AI Principles in 2019 (AIGO), Professor Hirano of Chuo University and I shared the key messages from the multi-stakeholder meetings that I chaired at the Council for Social Principles of Human-centric AI in Japan. Everyone who participated in these discussions felt confident that their outcomes will be an important initiative for implementing the OECD AI principles.

The topics discussed at the Council for Social Principles of Human-centric AI were Social Principles of Human-centric AI, Draft AI R&D guidelines for international discussions and Draft AI Utilization Principles. These projects first came to light at the Conference toward AI Network Society, MIC, Japan in 2017.

The Draft AI Utilization Principles, became the AI Utilization Guidelines in August 2019. These guidelines define AI issues and principles to follow around each issue. Quite naturally, the AI Utilization Guidelines have a lot in common with the OECD AI principles. Therefore, I think they can be labelled as an initiative that is closely related to the implementation of the OECD AI principles. In many places, the AI Utilization Guidelines refers directly to the OECD AI Principles. Here are two examples:

• “Consider for data used in AI learning and learning algorithms from the perspective of fairness, especially in the field of machine learning” – refers to the OECD AI Principle 1.2. Human-centered values and fairness
  
• “Procedures for ensuring AI explainability” refers to – the OECD AI Principle 1.3. Transparency and Explainability

The OECD AI principles promote “inclusive growth, sustainable development and well-being”. Japan wants to foster an AI economic society that is inclusive and where diverse players use AI and data, actively participate in social and economic activities, receive according to what they contribute, have a sense of fulfillment, have more leisure time, and so on. As such, Japan’s approach is consistent with the OECD AI principles.

National labour shares in each country have been on a downward trend for a long time, while distributions to intangible assets have been increasing. Much of the value created by data is distributed among intangible assets, suggesting companies that use data can be profitable, but public statistics are inadequate and the reality is not always clear. At the multi-stakeholder meetings where I have participated, the following items were discussed as central to sharing the value of data for AI in an objective way:

• Methods to measure the value of data that emphasizes versatility and reproducibility in each country and industry
• Policies to encourage data sharing
• Data ownership issues
• Fair distribution according to the effect and value of data

Tackling these issues will help to create a prosperous, trustworthy and beneficial AI economic society.