Speakers

• Dr Quentin Guibert — Centre de Recherche en Mathématiques de la Décision
• Marketa Krupova
• Arthur Charpentier
• Iris Nonneman
• Dr Simon Hatzesberger — Actuarial and Insurance Services, Deloitte, Munich, Germany
• Prof. Dr. Ralf Korn — University of Kaiserslautern
• Dr. David Atance — Universidad de Alcalá

Abstract

This article explores the potential of generative AI (GenAI) to support actuarial practice through four implemented case studies. It situates these case studies within the broader evolution of artificial intelligence in actuarial science, from early neural networks and machine learning to modern transformer-based GenAI systems. The first case study illustrates how large language models (LLMs) can improve claim cost prediction by extracting informative features from unstructured text for use in the underlying supervised learning task. The second case study demonstrates the automation of market comparisons using Retrieval-Augmented Generation to identify, extract, and structure relevant information from insurers’ annual reports. The third case study highlights the capabilities of fine-tuned vision-enabled LLMs in classifying car damage types and extracting contextual information from images. The fourth case study presents a multi-agent system that autonomously migrates actuarial legacy code from R to Python and validates the translation against the original code’s outputs. In addition to these case studies, we outline further GenAI applications in the insurance industry. Finally, we discuss the regulatory, security, dual-use and fraud, reproducibility, privacy, governance, and organisational challenges associated with deploying GenAI in regulated insurance environments.

Since the Lee and Carter (1992) model, many researchers have focused on forecasting age-specific mortality rates. Recent research highlights the use of mortality improvements to increase forecasting accuracy. We propose a mortality improvement model inspired by interest rate modeling, where the changes in the mortality curve depend linearly on a reduced number of age-specific or “key” age(s). These variables are directly observable and capture the general mortality trend. Previous versions used one key age, whereas we introduce a second factor to improve the explanatory and forecasting power. Across populations, the first key age is consistently around 85, while the second is near age 30. We compare the accuracy of the models with seven benchmark mortality models in six populations, including the pandemic period. The results show that including a second factor enhances performance only when it is correlated with a sufficiently wide range of age-specific rates.

Looking for more in the field?

Explore more events from Online Presentation of new EAJ issues and Springer Nature's Mathematics & Statistics.