Documentation is a vital component of any artificial intelligence (AI) model or system, forming the foundation for a robust governance framework. Comprehensive documentation is essential not only for compliance and regulatory purposes, but also for transparency, accountability of the AI systems and continuity of operations.
While this paper aims to assist actuaries in creating effective documentation for their AI models or systems, it makes no claim to be exhaustive. It outlines key elements that may be considered as good practice for documentation of an AI model or system. The paper covers the elements of documentation in all stages of the model lifecycle, from Data, through Model Development to Model Deployment. The level of detail included in AI model documentation should reflect proportionality and could vary depending on a number of factors, such as the level of significance, risk and complexity of the AI models. Actuaries are encouraged to use professional judgement and tailor the documentation to these factors, with the goal of meeting the needs of the end users of the documentation. In this document, the terms “AI system” and “AI model” align with those presented in the Artificial Intelligence Governance Framework paper.1 In summary, an AI system is an overall machine-based system that infers from inputs how to generate outputs, and an AI model is a core component within an AI system used to make such inferences.
