Governance Guidelines for Implementation of AI Principles

Governance Guidelines for Implementation of AI Principles Ver. 1.0 July 9, 2021 Expert Group on How AI Principles Should be Implemented AI Governance Guidelines WG Table of Contents A. INTRODUCTION 1. THE AIM OF THE AI GOVERNANCE GUIDELINES 2. LEGAL EFFECT OF THE GUIDELINES 3. RELATIONSHIP WITH OTHER GUIDELINES, ETC. 4. HOW TO USE THE AI GOVERNANCE GUIDELINES 5. LIVING DOCUMENT B. DEFINITIONS C. AI GOVERNANCE GUIDELINES 1. CONDITIONS AND RISKS ANALYSIS (1) Understanding positive and negative impacts that AI systems may have (2) Understanding social acceptance of AI system development and operation (3) Understanding your company's AI proficiency 2. GOAL SETTING (1) Considering setting AI governance goals 3. SYSTEM DESIGN (BUILDING AN AI MANAGEMENT SYSTEM) (1) Incorporating gap analysis between AI governance goals and current state to address the gap into the management system as essential process Ensuring consistency with industry-standard gap analysis processes Providing users with sufficient information on potential gaps and measures to address them Providing AI system developers with sufficient information for gap analysis (for data providers) (2) Improving literacy of human resources responsible for AI management systems (3) Reinforcing AI management through cooperation between companies/departments such as by proper information sharing Understanding the current state of information sharing between plural companies Encouraging gathering information and exchanging views routinely for conditions and risks analysis (4) Reducing incident-related burdens on users by preventing incidents and through early response Allocating burdens of addressing uncertainties among companies appropriately Considering in advance actions to take in response to incidents/disputes 4. IMPLEMENTATION (1) Ensuring readiness for explanation about implementation status of AI management systems (2) Ensuring readiness for explanation about operating status of individual AI systems (3) Considering proactively disclosing ranking implementation of AI governance as non-financial information 5. EVALUATION (1) Verifying an AI management system works appropriately (2) Considering seeking feedback from outside stakeholders 6. RE-ANALYSIS OF CONDITIONS AND RISKS (1) Re-implementing Action Targets 1-1 to 1-3 in a timely manner 1. The Aim of the AI Governance Guidelines In March 2019, Japan published the document called Social Principles of Human-centric AI adopted by the Integrated Innovation Strategy Promotion Council, which contributed to the formulation of the OECD recommendations on Artificial Intelligence. The document set social principles for AI that is to be implemented in the society as a whole, and it states that, based on such principles, companies involved in AI business typically as a developer and operator should establish and comply with the goals to be implemented according to the purpose and method of their AI business such as development and operation. The social principles for AI are comprised of seven principles: Human-centric, Education/Literacy, Privacy Protection, Ensuring Security, Fair Competition, Fairness Accountability and Transparency, and Innovation. The Governance Guidelines for Implementation of AI Principles present action targets to be implemented by an AI company, with the aim of supporting the implementation of the AI principles that is required for the facilitation of deployment of AI. Hypothetical examples of implementation corresponding to each of the action targets and practical examples for gap analysis between AI governance goals and current state are also shown. However, the examples of implementation and examples for gap analysis are for reference purposes only and are not intended to be exhaustive. 2. Legal Effect of the Guidelines The Guidelines themselves are not legally binding. The Guidelines are comprised of action targets to be implemented, examples of implementation, and examples for gap analysis, which summarize typical targets and practical examples that are shared in society to a certain extent. As the Social Principles of Human-centric AI, which are also not legally binding, are respected in society due to their universal nature, the Guidelines are also expected to support company voluntary efforts as widely shared material that is referred to by AI companies who are involved in AI business, typically the development and operation of AI systems, in their business transactions, and through the development of a common understanding among stakeholders on the implementation of AI principles. Please note that even if a company establishes structures in accordance with the Guidelines, it does not necessarily mean that the company complies with relevant laws, therefore please be mindful of complying with the relevant laws. 3. Relationship with other guidelines While new elements have been added, the Guidelines focus on incorporating essence of various codes, guidelines, assessment lists and other documents published in and outside Japan, with the aim of integrating the relevant guidelines. By cross-referencing a variety of relevant documents, the Guidelines are designed to serve as comprehensive guidelines as a whole. 4. How to Use the AI Governance Guidelines The Guidelines are comprised of the main part and Appendices. The action targets are general and objective ones that should be implemented by every AI company involved in AI business, typically the development and operation of AI systems that could have a certain level of negative impacts on society. The decision on whether to adopt examples of implementation and examples for gap analysis is left up to the discretion of the AI company, which may modify or select appropriate examples according to its circumstances. 5. Living Document AI technologies are in the midst of rapidly evolving. Society will accumulate knowledge to maximize positive impacts brought by AI systems, while managing negative impacts at a level acceptable to stakeholders. Therefore, it is unlikely that the Guidelines will adequately function in the future without revisions being made. For the improvement of AI governance, it is essential to continue reviewing how AI governance and the Guidelines should be and make necessary revisions through multi-stakeholder participation, while referring to the design concept of agile governance. This manual import source is derived from official METI PDF text exposed through the official web viewer for: https://www.meti.go.jp/shingikai/mono_info_service/ai_shakai_jisso/pdf/20210709_9.pdf