Overview
This specific report provides insights from senior actuarial leadership interviews to understand the top AI risks for actuaries, and emerging trends and non-obvious themes.
The Society of Actuaries (SOA) and the AI Risk Initiative at MIT FutureTech are collaborating on the Foresight Project, a multi-method and multi-phase research initiative on AI. This project examines current and emerging risks to the actuarial profession, to inform decisions about AI governance, professional education, and workforce development for the future.
Key Findings
Successful integration of AI into actuarial work requires a lot of human touch and understanding. The interviews of senior actuarial leadership provided insights on AI risks for actuaries. Key findings include:
- AI use is experimental and fragmented in the actuarial profession. Actuaries are often caught between top-down pressure to adopt AI but no practical roadmap.
- Top concerns include governance failure, transparency and interpretability challenges, fraud and scams, privacy, and model failures including hallucinations.
- Shared concerns include talent loss, displacement of actuaries, and loss of fundamental skills for human oversight and accountability.
- Leaders anticipate disruptive changes ahead with agentic and autonomous AI, and broader societal shocks that could destabilize insurance markets.
Acknowledgements
This project was funded by a gift from the Society of Actuaries (SOA) Research Institute. The authors thank Dale Hall and Ryan Silhavy at the SOA Research Institute for their valuable input with project planning, recruitment, and review, and for their support throughout this project. The authors are grateful to all senior actuarial leaders who contributed their time and insights to the interviews.
FAQ
This research examines current and emerging risks to the actuarial profession and how AI is reshaping the profession. The findings provide an evidence base for decisions about AI governance and workforce development to prepare the actuarial profession.
Yes, large language models were used to assist with report editing, cross-checking themes and quotes against original transcripts, and building a qualitative data analysis tool for organizing interview transcripts and for enabling thematic coding. All themes across interviews were coded manually.