國際精算學會(The International Actuarial Association, IAA) 近期發布了三篇由人工智慧工作小組(Artificial Intelligent Task Force, AITF)撰寫的論文，內容著重於設計、開發及使用人工智慧模型與系統時，應留意的相關風險。

1. Artificial Intelligence Governance Framework 

This paper provides foundational guidance on governance areas impacted by AI. It highlights key principles and best practices for managing risks related to data, modelling, and outcomes once AI systems are deployed. The paper is designed to strengthen actuaries’ understanding of governance responsibilities and support their oversight of AI within actuarial and broader financial contexts.

2. Testing of Artificial Intelligence Models 

This paper offers guidance on the principles and methodologies that underpin reliable and ethical AI model testing. It outlines a principle-based approach covering data creation, partitioning, assessment of fairness and explainability, evaluation of robustness and accuracy, and the implementation of continuous testing and monitoring throughout the model lifecycle. This paper underscores the actuary’s role in testing, supporting the validation of AI models and enhancing confidence in their use.

3. Documentation of Artificial Intelligence Models and Systems

This paper emphasizes the critical importance of clear, comprehensive and proportionate documentation throughout the entire model lifecycle. Recognizing documentation as the backbone of effective governance, the paper outlines essential elements needed to ensure transparency, accountability, regulatory compliance, and continuity of operations. It is intended for model developers, reviewers and validators who rely on consistent, well-structured documentation to support model risk management.

相關論文已發表在IAA官網>KNOWLEDGE/PAPERS，敬請參閱。