It’s 2030. Artificial intelligence (AI) has gained the competency to answer many actuarial questions that traditionally were answered exclusively by actuaries. Therefore, actuaries must develop new roles and acquire new skills. Paradoxically, AI has increased the work actuaries must do, as we end up with the new job of directing and validating the actuarial work of AI.
Two skills that have become especially important in the profession are (1) using one’s expertise to formulate critical questioning prompts to AI and (2) cultivating life-long learning based on a deep curiosity and thirst for new ideas. These enable the actuary to prompt AI more intelligently and find more truths in a complex world filled with uncertainty.
Our actuary, Sam, is waiting for AI to develop the cash flow testing report for IIC (The Intelligent Insurance Company). Sam has a standard protocol that he follows with important reports like this one. This protocol, based on his expertise and experience, involves three layers of questions followed by his conclusion about the report.
Layer 1 Questions
The first layer follows the Actuarial Standard of Practice (ASOP) No. 7 for Life and Health Cash Flow Testing. Sam discovered a couple of years ago that the ASOPs can generally be turned inside out. Instead of using it to decide how to do a project like cash flow testing, he can use the options that the standard grants to the practicing actuary and turn them into questions. The first layer is all about minimally satisfying what is explicitly stated in the ASOP.
He thinks of the questions as falling under four major categories: data, assumptions, methods and conclusions. Here are some examples of potential questions based on the ASOP:
- Data: If data was provided by a third party, has it been rigorously reviewed for baseline reasonableness and consistency before being fed into the projection? (3.8a)
- Assumptions: Are the interdependent assumptions mutually consistent within each scenario, or does a change in one economic factor create a logical contradiction in another? (3.6)
- Methods: Does this specific methodology properly account for the material variations in asset and liability cash flows across a wide range of plausible scenarios, or does it oversimplify them? (3.3(a) and (b))
- Conclusion: Does the final set of scenarios capture a wide enough range and number of outcomes to reasonably represent the underlying economic variability and tail risks? (3.6)
These should be easy for the AI model to answer. The ASOP was part of the special training data for the model, and it was specifically instructed to perform a cash flow test that would conform with both the ASOP and the cash flow testing model law. Layer 1 questions are perfunctory. They need to be asked and the responses logged.
Layer 2 Questions
Now Sam can move on to Layer 2 questions, which are ASOP adjacent. They are not included in the ASOP, but they could have been. Layer 2 questions mostly come from personal experience with cash flow test work in previous years, are a result of the different situations experienced or are the questions Sam’s boss frequently asked after Sam had completed the report.
Examples of these types of questions include the following:
- Data: We matched data from two systems for the liability cash flows. Last year, we installed a new claims system, and some data elements like the claims paid date are now consistent with the actual payment, whereas in prior years, they were sometimes estimates. How does that discontinuity of data impact the projections?
- Assumptions: How has the impact of the new agent contract, especially renewal commissions, been reflected in the assumptions of contract persistency?
- Methods: Our execution of asset trades is usually bunched; we like to go out to the market with a certain minimum number of transactions at a time. Does the model reflect that bunching, or are transactions averaged?
- Conclusion: The conclusion of reserve adequacy seems to rest in part upon execution of frequent rebalancing. When is the last time we actually rebalanced, and were those transactions consistent with the protocol assumed in the report?
Deep Curiosity
Often Sam stops there. He has done the minimum with Layer 1 questions and gone the extra mile with Layer 2. But he knows there are more questions that he needs to ask. This year, things are pretty chaotic, with market volatility up to nearly record levels and the new agent commission contract messing with the future persistency expectations.
These additional questions go beyond the usual actuarial tool kit. They come from lifelong learning driven by a deep curiosity and thirst for new ideas. Before we see how Sam applies this to his situation, let’s first discuss the necessary motivational driver—deep curiosity.
I like to think of three levels of curiosity: idle curiosity, which I equate with channel surfing on cable or streaming services; normal curiosity, which is what causes gaper delays on the highway; and deep curiosity, which is this motivating curiosity.
Deep curiosity is a concept that appears in psychology, philosophy and education. It can be like physical hunger, and studies since 2014 have shown a physical link to the same neurotransmitter pathways.[1] An individual feels compelled to make a sustained and intense effort to fully understand some topic.
For Sam, deep curiosity was triggered by the advent of AI itself and an interest in understanding the possible ways that human cognition might differ from machine cognition. Deep curiosity is needed to create the special knowledge that leads to divergent lines of inquiry, and it is also the motivation for bothering to pursue Level 3 questioning of an AI answer.
Layer 3 Questions
Sam has been deeply curious for the past three or four years. He has been studying various ways that people think about the problems of everyday life. He has yet to find a single source, a place to go for this type of information. It seems that just as there is a standard way that actuaries think and solve problems, there are standard approaches that other professionals use. Here are some examples that Sam has found:
Systems thinking (interconnections and feedback loops): reveals human feedback loops; maps tipping points; analyzes nonlinear system characteristics. Use when major behavioral parameters are fixed but the situation is likely to trigger reflexive changes in actions.
- Systems question: If we lower our crediting rates to preserve cash flow under this scenario, how will the reactions of our agents and market competitors create a feedback loop that diverges sharply from our static lapse assumptions?
Plural rationalities thinking (contrasting worldviews): identifies competing stakeholder worldviews and surfaces conflicting priorities. Use when a decision requires genuine buy-in from parties whose core values or incentives are structurally in conflict.
- Plural rationalities question: How does this proposed rate or dividend structure fall apart when stress-tested against the contrasting worldviews of our risk-averse compliance officers, our growth-oriented sales heads and defensive consumer advocacy groups?
Ethical reasoning (fairness, justice and trust): tests moral defensibility of optimization outcomes. Use when a first answer heavily penalizes a specific segment of the population.
- Ethical question: Looking strictly at the distributions of risk in this model, which specifies a disproportionate burden on certain groups in our policyholder base, is that distribution morally defensible in the public eye?
Creative thinking (imagination and adaptability): generates novel structural alternatives unconstrained by precedent. Use when a regime shift or other discontinuity renders historical data paths broken or irrelevant.
- Creative thinking question: If we relax the constraint that our asset-liability matching must conform to our historical investment policy, what alternative structures might address this unprecedented risk scenario?
Economic thinking (incentives, resource constraints and opportunity cost): focuses on the rational behaviors, competitive frictions and capital trade-offs that drive market participants. Use when AI assumes that altering one operational lever will leave all other stakeholder behaviors completely static.
- Economic question: While this strategy optimizes our projected financial margin on paper, how will our distribution channels and market competitors shift their reactions to their economic incentives to capture optimal margin for themselves?
Sam has collected these ideas over several years out of curiosity, but this is not his full list. He finds that adding questions from these five different thinking styles usually adds one or several good questions that will cause AI to struggle to answer.
What he is doing with those five alternate thinking approaches is deliberately stepping “outside the box” of the standard actuarial thinking approach. There are many ways to step outside the normal box. It could involve talking to someone who works in a totally different field. The thinking needed to reframe the question in order to pose it to someone else is often enough to get you to step outside of the box.
Back to 2026
This story of 2030 is not a prediction; it is a scenario we have tested. In that test, critical questioning and deep curiosity are valuable skills for actuaries. They allow actuaries to have a valuable role even if AI takes over much of the work we do right now. AI may not become that effective at doing actuarial work, but the economics of knowledge work is fundamentally changing with AI. Since the machine is automating the production of answers, we can’t compete with it by trying to out-calculate it.
Deep curiosity for new ideas and critical questioning prompts are new pipelines and professional skills through which human expertise can be injected into the workflow. By transforming our traditional expertise into an active system of inquiry—combining defensive cross-examination with offensive, curiosity-driven discovery—we stop waiting to be replaced and instead become the active directors of an AI force multiplier.
The key takeaway is recognizing that exploration driven by deep curiosity and intelligent prompts arising from our expertise and experience is now a core professional skill, not a soft one.
This article is provided for informational and educational purposes only. Neither the Society of Actuaries nor the respective authors’ employers make any endorsement, representation or guarantee with regard to any content, and disclaim any liability in connection with the use or misuse of any information provided herein. This article should not be construed as professional or financial advice. Statements of fact and opinions expressed herein are those of the individual authors and are not necessarily those of the Society of Actuaries or the respective authors’ employers.
Dave Ingram, FSA, CERA, is a researcher, writer and part-time consultant on risk and risk management. He serves as a member of the board of the Society of Actuaries (2021–2027) and is also an editor of the new SOA Research Institute’s Actuarial Intelligence Bulletin. Dave can be reached at daveingram@optonline.net.
Endnote
[1] Matthias J. Gruber, Bernard D. Gelman and Charan Ranganath, “States of Curiosity Modulate Hippocampus-Dependent Learning via the Dopaminergic Circuit,” Neuron 84, no. 2 (2014): 486–96.