Claim your CPD points
Earlier this month, OpenAI shared more than 700 AI-generated mathematical manuscripts , including proposed solutions to longstanding research problems. The release sparked debate about how to verify the proofs, establish their novelty and assess their significance. OpenAI subsequently committed US$5 million to a Simons Foundation initiative supporting mathematicians in examining and building on AI-generated research.
I find this fascinating. AI can produce sophisticated mathematical work in volumes that place enormous demands on human experts. The bottleneck is shifting from producing results to having the expertise and judgement to evaluate and build on them.
I am experiencing something similar as an actuarial educator. My headache has shifted from too little, too late, to too much, too fast. Students can now produce far more work, but reviewing it raises different questions: Is the analysis correct? Are the assumptions reasonable? Does the student understand it? And are we even solving the right problem?
AI is making it easier to generate work, but harder to evaluate it.
Professional judgement and discernment develop through hands-on experience, tackling challenging problems, testing ideas, making mistakes and learning from the outcomes. If AI increasingly takes over that work, how will junior actuaries develop the expertise needed to evaluate it? And where will our next generation of experienced actuaries and leaders come from?
In August, ABC News reported the case of an Australian patient whose AI-assisted medical documentation falsely stated that she had taken psychedelic mushrooms. The error caused considerable distress and concern about possible consequences for her workers' compensation case. The specialist subsequently apologised and corrected the record.
Australia's health practitioner regulator, Ahpra, emphasised that clinicians remain responsible for checking AI-generated documentation. In this case the ability to generate records quickly was not matched by the ability to evaluate accuracy and ensure mistakes were corrected.
The lesson extends beyond healthcare. AI can produce outputs at remarkable speed, but professionals remain responsible for ensuring they are reliable and appropriate.
What does this mean for actuaries?
Actuarial education has traditionally focused heavily on generating answers: solving mathematical problems, building models, writing code, analysing data and preparing reports.
Of course, actuaries have always evaluated their work. But AI is changing the balance. As AI takes on more technical production, critically evaluating its outputs becomes an increasingly important part of our expertise.
I sometimes compare this to the difference between a student and a teacher. A student produces an answer. A teacher must determine whether it is correct, whether the reasoning is sound, what is missing and how it could be improved.
Consider an AI-generated insurance pricing model. It may have excellent predictive accuracy, sophisticated code and convincing explanations. But as actuaries, we need to ask three questions:
A technically impressive model is not necessarily a good model. And an accurate prediction does not automatically justify a decision.
These questions require deep technical knowledge, experience and professional judgement. We cannot critically evaluate what we do not understand.
Strong actuarial foundations therefore remain essential. AI does not eliminate the need for technical expertise. In many ways, it makes that expertise more valuable.
But good judgement alone is not enough. Actuaries also need to communicate their findings, collaborate across disciplines and learn to lead AI agents. These interpersonal and leadership skills will become increasingly important.
AI allows us to generate work at a scale that traditional review processes were never designed to handle. If one actuary can produce the analysis previously undertaken by an entire team, can that same actuary meaningfully evaluate everything?
Simply placing a human at the end of an automated process to approve the results is not enough. We need better ways to validate AI-generated work, identify decisions requiring closer scrutiny and recognise when recommendations should be challenged or rejected.
AI can assist with evaluation, but professional accountability remains.
This is where I see an opportunity for actuaries. Our profession combines technical rigour with judgement, ethics and responsibility for managing uncertainty. These strengths put us in a good position to help organisations adopt AI responsibly.
As AI changes the way we work, we need to rethink how we develop and maintain professional judgement and discernment. This applies to actuaries at every career stage, from graduates to experienced professionals.
A paper by Shaw and Nave discusses cognitive surrender: accepting AI-generated answers without sufficient independent scrutiny. This worries me as an educator.
I often describe my classroom as a cognitive gym. Just as we exercise our bodies to build physical strength, students need to exercise their minds to develop critical thinking and judgement.
In this gym, I want students to tackle challenging problems in teams, question assumptions, learn from mistakes and defend their decisions. They also practise evaluating AI-generated work, identifying errors, testing assumptions and comparing alternatives. Perhaps this applies to all of us, not just students. We need to visit the cognitive gym regularly to keep our thinking muscles strong and avoid cognitive surrender as we increasingly rely on AI.
I have been developing this approach through my Actuarial Datathon challenges , where students work together on real-world problems, including pet insurance pricing and travel insurance growth strategies. Multiple industry partners provide feedback and guidance, helping students understand practical constraints, challenge their approaches and improve their recommendations.
We also need both AI-free and AI-enabled learning environments. Students must develop foundational skills independently without AI, while also learning to work creatively and critically with AI.
As AI redefines entry-level roles, universities must also rethink the university experience. We may need to provide more of the practical experience graduates traditionally gained in their early careers, helping them develop the judgement these changing roles demand.
But universities cannot do this alone.
My call to the actuarial industry is to work more closely with universities to co-develop the next generation of actuaries, ready to thrive in an AI-enabled world. Industry professionals can help shape real-world challenges, mentor students and challenge their decisions. Universities can provide opportunities to experiment, make mistakes and develop judgement before entering the workforce.
After all, AI may be happy to do the homework, but someone still needs to know how to mark it. And that is a responsibility we cannot afford to outsource.
The views expressed in this article are those of the author(s) or working group named below, and do not necessarily reflect the views of the Actuaries Institute. This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivatives CC BY-NC-ND Version 4.0.
From model governance to machine learning, explore actuarial perspectives on data science and artificial intelligence
Subscribe to Actuaries Digital for free and receive the latest actuarial analysis, research, and commentary direct to your inbox