Life Conference 2024 C1: How to validate AI models – an actuarial perspective and optimising and automating controls: Lessons from banking
- Model Validation will become increasingly important for Insurers to undertake with indications from the PRA in the Dear CEO Letter and 2023 Business Plan that life insurers should consider how Model Risk Management proposals applying to the banking sector can apply to them. The government has also issued ‘AI regulatory principles’ and initial guidance on how these will be implemented by regulators. This presentation will aim to address the following:
• An overview of some common AI and Machine Learning (ML) models used in the insurance and finance sectors and the differences these may have to ‘standard’ insurance models.
• How to shape the validation framework for AI/ML models. This will consider how to use the Data Science Lifecycle for setting up governance and validation, the complexity and uncertainty guidelines around AI and ML, and how the scope may differ to a ‘standard’ insurance validation (e.g. consideration of ethics).
• A compare and contrast between ‘standard’ insurance validations and AI/ML validations. This would be broken down into the key components of a validation e.g. data, assumptions, documentation, output and results, model uses etc.
• A high level summary of common themes and findings when validating AI/ML models
Speakers Martin Hall and Diana Dobre, KPMG