« 米国 上院議員がデータ保護法案を再提出 | Main | 人工知能技術に芸術と人文科学はどのように貢献できるか? by John Tasioulas(オックスフォード大学のAI倫理研究所所長) »

2021.06.22

欧州保険職業年金局 (EIOPA) 欧州保険セクターにおける倫理的で信頼できるAIガバナンス原則に関するレポートを公表

こんにちは、丸山満彦です。

欧州保険職業年金局 (EIOPA) が欧州保険セクターにおける倫理的で信頼できるAIガバナンス原則に関するレポートを公表していますね。。。

 

European Insurance and Occupational Pensions Authority; EIOPA

・2021.06.17 EIOPA publishes report on artificial intelligence governance principles

・[PDF] 人工知能ガバナンス原則。TOWARDS ETHICAL AND TRUSTWORTHY ARTIFICIAL INTELLIGENCE IN THE EUROPEAN INSURANCE SECTOR - A report from EIOPA's Consultative Expert Group on Digital Ethics in insurance

20210622-22458

原則

Principle of proportionality
比例性の原則
Principle of fairness and non-discrimination 公平性と無差別の原則
Principle of transparency and explainability 透明性と説明可能性の原則
Principle of Human Oversight 人間による監督の原則
Principle of data governance of record keeping 記録管理のデータガバナンスの原則
Principle of Robustness and Performance 堅牢性とパフォーマンスの原則
   
   
Principle of proportionality: Insurance firms should conduct an AI use case impact assessment in order to determine the governance measures required for a specific AI use case. The AI use case impact assessment and the governance measures should be proportionate to the potential impact of a specific AI use case on consumers and/or insurance firms. Insurance firms should then assess the combination of measures put in place in order to ensure an ethical and trustworthy use of AI. 比例性の原則:保険会社は、特定のAIユースケースに必要なガバナンス対策を決定するために、AIユースケースの影響評価を行うべきである。AIユースケースの影響評価とガバナンス対策は、特定のAIユースケースが消費者や保険会社に与える潜在的な影響に比例すべきである。そして、保険会社は、倫理的で信頼できるAIの使用を保証するために導入された措置の組み合わせを評価すべきである。
Principle of fairness and non-discrimination: insurance firms should adhere to principles of fairness and non-discrimination when using AI. They should take into account the outcomes of AI systems, while balancing the interests of all the stakeholders involved. As part of their corporate social responsibility insurance firms should also take into account financial inclusion issues and consider ways to avoid reinforcing existing inequalities, especially for products that are socially beneficial. This includes assessing and developing measures to mitigate the impact of rating factors such as credit scores and avoiding the use of certain types of price and claims optimisation practices like those aiming to maximise consumers’ “willingness to pay” or “willingness to accept”. Fair use of data means ensuring that it is fit for purpose and respect the principle of human autonomy by developing AI systems that support consumers in their decision-making process. Insurance firms should make reasonable efforts to monitor and mitigate biases from data and AI systems. This may include using more explainable algorithms or developing fairness and non-discrimination metrics in high-impact AI applications. Insurance firms should develop their approach to fairness and keep records on the measures put in place to ensure fairness and non-discrimination. 公平性と無差別の原則:保険会社は、AIを使用する際、公平性と無差別の原則を遵守すべきである。保険会社は、関係するすべての利害関係者の利益のバランスを取りながら、AIシステムの結果を考慮するべきである。企業の社会的責任の一環として、保険会社は、特に社会的に有益な商品について、金融包摂の問題を考慮し、既存の不平等を強化しない方法を検討する必要がある。これには、クレジットスコアなどの格付け要素の影響を緩和するための評価と対策の策定、消費者の「支払い意思」や「引き受け意思」を最大限に引き出すことを目的とした、ある種の価格や保険金請求の最適化手法の使用を避けることなどが含まれる。データの公正な使用とは、データが目的に適合していることを保証し、消費者の意思決定プロセスをサポートするAIシステムを開発することで、人間の自律性の原則を尊重することを意味する。保険会社は、データやAIシステムからのバイアスを監視し、緩和するための合理的な努力をすべきである。これには、より説明可能なアルゴリズムを使用することや、影響力の大きいAIアプリケーションにおいて公平性と無差別性の評価基準を開発することが含まれる。保険会社は、公平性へのアプローチを開発し、公平性と無差別を確保するために導入された措置に関する記録を残すべきである。
Principle of transparency and explainability: Insurance firms should adapt the types of explanations to specific AI use cases and to the recipient stakeholders. Insurance firms should strive to use explainable AI models, in particular in high-impact AI use cases, although, in certain cases, they may combine model explainability with other governance measures insofar as they ensure the accountability of firms, including enabling access to adequate redress mechanisms. Explanations should be meaningful and easy to understand in order to help stakeholders make informed decisions. Insurance firms should transparently communicate the data used in AI models to consumers and ensure that they are aware that they are interacting with an AI system, and its limitations. 透明性と説明可能性の原則:保険会社は、説明の種類を特定のAIユースケースと受容者のステークホルダーに合わせるべきである。保険会社は、特にインパクトの大きいAIユースケースでは、説明可能なAIモデルを使用するよう努めるべきであるが、場合によっては、適切な救済メカニズムへのアクセスを可能にするなど、会社の説明責任を確保する限りにおいて、モデルの説明可能性を他のガバナンス手段と組み合わせることができるものとする。説明は、ステークホルダーが十分な情報を得た上で意思決定できるように、意味があり、理解しやすいものでなければならない。保険会社は、AIモデルに使用されているデータを消費者に透明に伝え、消費者がAIシステムと対話していることとその限界を認識していることを確認すべきである。
Principle of Human Oversight: Insurance firms should establish adequate levels of human oversight throughout the AI system’s life cycle. The organisational structure of insurance firms should assign and document clear roles and responsibilities for the staff involved in AI processes, fully embedded in their governance system. The roles and responsibilities of staff members may vary from one AI use case to another. It is also important that insurance firms assess the impact of AI on the work of employees and provide staff with adequate training. 人間による監督の原則:保険会社は、AIシステムのライフサイクルを通じて、適切なレベルの人間による監督を確立すべきである。保険会社の組織構造は、ガバナンス・システムに完全に組み込まれた形で、AIプロセスに関わるスタッフの明確な役割と責任を割り当て、文書化すべきである。スタッフの役割と責任は、AIのユースケースごとに異なる場合がある。また、保険会社は、AIが従業員の業務に与える影響を評価し、スタッフに適切なトレーニングを提供することが重要である。
Principle of data governance of record keeping: The provisions included in national and European data protection laws (e.g. GDPR) should be the basis for the implementation of sound data governance throughout the AI system lifecycle adapted to specific AI use cases. Insurance firms should ensure that data used in AI systems is accurate, complete and appropriate and they should apply the same data governance standards regardless of whether data is obtained from internal or external sources. Data should be stored in a safe and secured environment and, in particular for high-impact use cases, insurance firms should keep appropriate records of the data management processes and modelling methodologies in order to enable their traceability and auditability. 記録管理のデータガバナンスの原則:国内および欧州のデータ保護法(例:GDPR)に含まれる規定は、特定のAIユースケースに適応したAIシステムのライフサイクルを通じて健全なデータガバナンスを実施するための基礎となるべきである。保険会社は、AIシステムで使用されるデータが正確で、完全で、適切であることを保証すべきであり、データが内部または外部ソースから取得されるかどうかにかかわらず、同じデータガバナンス基準を適用すべきである。データは、安全でセキュアな環境に保管されるべきであり、特に影響の大きいユースケースについては、保険会社は、データ管理プロセスやモデリング手法の適切な記録を残し、そのトレーサビリティと監査可能性を確保すべきである。
Principle of Robustness and Performance: Insurance firms should use robust AI systems, both when developed in-house or outsourced to third parties, taking into account their intended use and the potential to cause harm. AI systems should be fit for purpose and their performance should be assessed and monitored on an on-going basis, including the development of relevant performance metrics. It is important that the calibration, validation and reproducibility of AI systems is done on a sound manner that ensure that the AI systems outcomes are stable overtime and/or of a steady nature. AI systems should be deployed in resilient and secured IT infrastructures, including against cyber-attacks. 堅牢性とパフォーマンスの原則:保険会社は、自社で開発する場合も、第三者に委託する場合も、その使用目的と被害をもたらす可能性を考慮した上で、堅牢なAIシステムを使用すべきである。AIシステムは、目的に適合したものでなければならず、その性能は、関連する性能指標の開発を含めて、継続的に評価・監視されるべきである。AIシステムの校正、検証、および再現性は、AIシステムの結果が安定した時間経過および/または安定した性質であることを保証する健全な方法で行われることが重要である。AIシステムは、サイバー攻撃に対するものも含め、回復力があり安全なITインフラに導入されるべきである。

 

 

I. FOREWORD I. 序文
II. EXECUTIVE SUMMARY II. エグゼクティブ・サマリー
III. GOVERNANCE PRINCIPLES FOR AN ETHICAL AND TRUSTOWOR- THY AI IN THE EUROPEAN INSURANCE SECTOR III. 欧州の保険部門における倫理的で信頼できるAIのためのガバナンス原則
IV. INTRODUCTION IV. 序論
1. The use of AI in the insurance sector 1. 保険分野におけるAIの利用について
2. The importance of digital ethics in insurance 2. 保険におけるデジタルエシックスの重要性
3. Approach taken, definitions and scope 3. 取り組み方、定義、範囲
V.  AI USE CASE IMPACT ASSESSMENT V.  AIユースケースの影響評価
1. Assessing the impact of an AI use case to determine the relevant governance measures 1. 関連するガバナンス対策を決定するために、AIユースケースの影響を評価すること
2. AI use case impact assessment framework 2. AIユースケースの影響評価フレームワーク
VI.  FAIRNESS AND NON-DISCRIMINATION VI.  公平性と無差別
1. Fairness and non-discrimination in insurance 1. 保険における公平性と無差別性
2. Fairness and non-discrimination in specific AI use cases in insurance 2. 保険における特定のAIの使用例における公平性と無差別性
VII. TRANSPARENCY AND EXPLAINABILITY VII. 透明性と説明可能性
1. Transparency and explainability in the insurance sector 1. 保険分野における透明性と説明可能性
2. Transparency and explainability in specific AI use cases in insurance 2. 保険における特定のAI使用例における透明性と説明可能性
VIII. HUMAN OVERSIGHT VIII. 人間による監視
1. Human oversight in the insurance sector 1. 保険分野における人間による監視
2. Human oversight in specific AI use cases in insurance 2. 保険における特定のAI使用例における人間による監視
IX.  DATA GOVERNANCE AND RECORD KEEPING IX.  データガバナンスと記録保持
1. Data governance and record keeping in insurance 1. 保険におけるデータガバナンスと記録の保持
2. Data governance and record keeping in specific AI use cases in insurance 2. 保険における特定のAI使用例におけるデータガバナンスと記録保持
X.  ROBUSTNESS AND PERFORMANCE X.  堅牢性とパフォーマンス
1. Robustness and performance in insurance 1. 保険における堅牢性とパフォーマンス
2. Robustness and performance in specific AI use cases in insurance 2. 保険における特定のAI使用例での堅牢性と性能
XI.  CONCLUSION XI.  おわりに
ANNEX 附属書
ANNEX 1 – BENEFITS AND ETHICAL CHALLENGES OF AI USE CASES ACROSS THE INSURANCE VALUE CHAIN 附属書1 - 保険のバリューチェーンにおけるAI使用事例の利点と倫理的課題
ANNEX 2 – AI USE IMPACT ASSESSMENT FRAMEWORK 附属書2 - AI使用の影響評価フレームワーク
ANNEX 3 – NON DISCRIMINATION REGULATORY FRAMEWORK IN INSURANCE 附属書3 - 保険における非差別的な規制の枠組み
ANNEX 4 – LIST OF ACRONYMS 附属書4 - 頭字語リスト
ANNEX 5 – MEMBERS OF EIOPA’S CONSULTATIVE EXPERT GROUP ON DIGITAL ETHICS IN INSURANCE 附属書5 - 保険におけるデジタル倫理に関するEIOPAの諮問専門家グループのメンバー
GLOSSARY 用語集
REFERENCES 参考文献

 

■ 参考

●まるちゃんの情報セキュリティ気まぐれ日記

AIガバナンス関係

・2021.02.15 経済産業省の「我が国のAIガバナンスの在り方 ver. 1.0(中間報告書)」に対する経団連等の意見

2020.11.10 中国 TC260 パブコメ AI倫理に関するガイドライン案 

・2020.10.12 欧州議会は「人工知能、ロボットおよび関連技術の倫理的側面の枠組み」を採択しましたね。。。

・2020.03.31 AI 倫理指針の動向とパーソナル AI エージェント by 中川裕志先生   AI 原則は機能するか?―非拘束的原則から普遍的原則への道筋 by 新保史生先生

・2020.03.28 EU Ethics Guidelines for Trustworthy AI(信頼できるAIのためのEU倫理ガイドライン)

 

AI関係全般

 

|

« 米国 上院議員がデータ保護法案を再提出 | Main | 人工知能技術に芸術と人文科学はどのように貢献できるか? by John Tasioulas(オックスフォード大学のAI倫理研究所所長) »

Comments

Post a comment



(Not displayed with comment.)


Comments are moderated, and will not appear on this weblog until the author has approved them.



« 米国 上院議員がデータ保護法案を再提出 | Main | 人工知能技術に芸術と人文科学はどのように貢献できるか? by John Tasioulas(オックスフォード大学のAI倫理研究所所長) »