What are the biggest challenges of using artificial intelligence in actuarial science?

Data quality is essential in AI for actuarial science. Inaccurate or biased data can undermine model effectiveness. Actuaries must ensure rigorous data validation and governance to maintain reliability and alignment with business goals and regulation.

Yes but that applies to any technology in actuarial science. GIGO. This is not the first technology to fall foul of poor data quality in insurance

La IA está revolucionando la ciencia actuarial, pero viene con sus desafíos: Datos: Necesitamos datos buenos. Si son incorrectos, las predicciones también lo serán. Entender la IA: A veces es complicado saber cómo la IA llega a sus conclusiones. Reglas y ética: La IA debe usarse de manera justa y cumplir con las leyes. Evolución de Roles: Los actuarios deben adaptarse, aprendiendo nuevas habilidades. Integración: Incorporar IA sin causar un lío en sistemas existentes. Educación: Hay que estar al día con tecnología y análisis de datos. La IA promete hacer nuestro trabajo más eficiente, desde evaluar riesgos hasta tomar decisiones. Es como aprender a manejar un gadget nuevo: al principio cuesta, pero luego no puedes vivir sin él.

Ethics and AI is a fresh new topic (and great to see the current Lord Mayor of London is promoting a course on the topic). Different actuarial roles need to consider this vert differently so the first step is to understand the role of ethics in your role generally, and then add on the AI questions. For example those working in the B2C pricing space need to consider it far more than an actuary working in reserving for a commercial insurer or broker.

Following a fascinating presentation from Prof Stephen Roberts at TINtech London Market I suspect we will very soon be discussing bias in synthetic data, which many see as a potential solution to both poor data quality, but also the security and privacy issues around using “real data”

I think a key issue here is that actuarial science is a vague and broad term. The ethics of AI in different disciplines is completely different. The ethical questions in pensions are different to the ethics in general insurance. So one ends up with quite meaningless phrases that could honestly be applied to any ethical question whether tangentially related to actuarial science or not. It is much more meaningful to ask questions like for example what are the ethical questions of valuing a pension scheme using AI or what is the ethics of using AI to set general insurance prices. Rather than categorizing these things under the banner term actuarial science. The two activities are utterly different in almost every way except they both use maths

Ethics in AI is a profound and tricky topic that we will be working on for years. Regulatory compliance ought to be easier - it is just necessary to ensure it is considered from the start rather than introducing it once tech and data teams are already producing pilots or live AI tools. It's another reason why AI should be seen as a topic that cuts across the company from the very start rather than being incubated deep within the IT or Data teams.

AI in actuarial science faces regulatory hurdles like GDPR, anti-discrimination laws, and Solvency II. The EU AI Act adds further compliance requirements. Local guidelines, such as DAV in Germany, are also crucial.

I agree with Paolo here: the most important thing is to give it a go! Perhaps not in a live environment, but doing nothing and seeing how it evolves before copying the leaders is unlikely to work in this case. They’ll be doing something much better by the time you’ve caught up.

Two thoughts here - firstly, there are few experts in this area - we are all learning. So people should not be put off by worrying they are not experts. Getting going is the most important step. Second point is that we are very early in the journey. Much of the oversight and learning now is on quite basic and banal stuff but it builds the framework and the foundations for oversight on far more sophisticated AI activities in the year head. Thus, again, this is about getting going as soon as possible rather than waiting for a later stage, at which point the challenge of starting from nothing will be far harder.

Using AI to set general insurance prices has more in common with the question of using AI to determine general insurance claims payouts made to customers than it does to other areas of actuarial science like life assurance or pensions. Or even other areas of general insurance like reserving and capital/solvency modelling. The regulation requirements are different as are the impacts on people. Lumping them together and then trying to develop human oversight frameworks for all of these diverse sectors at once leads to saying meaningless phrases that apply everywhere like 'monitor and audit the AI models'.

I would strongly suggest actuarial leaders to encourage EVERYONE in their teams to start engaging with AI in different ways. Everyone acts in different ways - some would love to go to evening MeetUps, other would prefer to get on external mailing lists, yet others are happy and able to summarise white papers from consulting firms etc. Even though this topic is still small it will grow and as I always say, you can't be a fast-follower from a standing start.

For AI not to mean "Automated Ignorance" it is important that Actuaries (and other users) keep line of sight as to what the output is, and why it is like that. No actuary will escape sanction for recommendations based on "but the computer said..." The "intelligence" dimension only emerges as the algorithms used for machine learning become sophisticated enough to deal with left-field input or deceptions. Bias and error can otherwise be swiftly embedded in the process, resulting in poor decisions and structural inequity. AI has the look of .Com from the late 90s. There will be bubbles that burst, and some players that will emerge as powerhouses. Others will boom and bust.

I recommend that everybody become familiar with AI tools. AI has made everybody a programmer, and it would be good to understand how to use the tools for daily tasks.