Actuarial Notes

A Primer on Generative AI for Actuaries

By Stephen Carlin and Stephan Mathys · Society of Actuaries Research Institute · 2024

Study guide: Exam CP-312

A February 2024 report of the Society of Actuaries Research Institute on generative artificial intelligence, which builds on the core concepts of AI to generate new data or material such as text, images or computer code. It surveys where generative AI may emerge in actuarial work (general task automation, coding and model documentation, data enrichment, scenario development, claims and underwriting), the limitations actuaries must be cognizant of, and the architecture and deployment considerations before putting an application into a production process.

On the syllabus

  • Exam CP-312 — Sections 1-5, for outcomes 2a, 2b and 2c (Topic 2).

Executive Summary

1 Introduction

2 An Introduction to Generative AI

  • 2.1 Overview of Generative AI
  • 2.2 Potential Benefits of Generative AI for Actuaries
  • 2.3 Examples of Where AI Can Bring Significant Benefits:

3 Applications of Generative AI for Actuaries

  • 3.1 Generative AI for General Productivity
    • 3.1.1 Meetings
    • 3.1.2 Summarizing Documents
    • 3.1.3 Learning Tools
    • 3.1.4 Drafting Content
  • 3.2 Generative AI for Coding and Software Development
    • 3.2.1 Copilot Tools
    • 3.2.2 Code Conversion
    • 3.2.3 Building Actuarial Models
  • 3.3 Model Documentation and Governance
  • 3.4 Enriching, Manipulating and Analyzing Data
    • 3.4.1 Enriching Data
    • 3.4.2 Manipulating Data
    • 3.4.3 Data Analysis
  • 3.5 Scenario Analysis
  • 3.6 Automation and Efficiency
    • 3.6.1 Automation
    • 3.6.2 Efficiency
  • 3.7 Claims
  • 3.8 Underwriting

4 Practical Application of Generative AI

  • 4.1 Limitations and Areas of Concern When Using AI
  • 4.2 Architectural and Deployment Considerations
    • 4.2.1 Models and Platforms
    • 4.2.2 Training Process and Data
    • 4.2.3 Prompt Engineering, Transfer Learning and RAG
  • 4.3 Checklist for Using Generative AI

5 Summary (Conclusions)

6 Acknowledgments

Appendix A Summary of Key GenAI Models and Techniques

Appendix B Glossary of Terms

References

About The Society of Actuaries Research Institute

Sources