AI: OK?

Artificial Intelligence (AI) has been a hot topic for a long time now, and is a development which promises to further transform many aspects of our lives. To understand what this could mean for actuaries, we should start from something we are certain of: data sets are the foundation for predictions which underpin risk assessment. Machine learning and other AI tools make it possible to consider vastly expanded data sets and, as such, to potentially develop entirely new ways to evaluate and analyse risks.

Written on 02/21/23

written by Esko Kivisaari
Chair AAE Artificial Intelligence-Data Science Working Group

Many people – in insurance, technology and elsewhere – instinctively feel that more information can only be a benefit. And studies suggest that, so far, that has been the case. But others worry about problems downstream: what if AI facilitates so much granularity of data that the pooling of similar risks becomes practically impossible? Or uncertainty is eliminated to such an extent that risk-sharing is no longer appropriate? Perhaps predictions could become so precise that premiums will simply be unaffordable? Or perhaps greater intelligence in risk calculation could even change people’s behaviour? These considerations are key to the continued success of a 'mutualised' approach to risk. And actuaries would also interject that greater accuracy in predictions doesn’t change the fact that actual outcomes will always be stochastic.

On the other hand, harnessing AI in relation to certain admin-intensive, high-frequency risks is likely to make insurers able to offer cover that is currently uneconomic for their clients, thereby expanding insurability. Furthermore, many actuarial models rely on calibration by humans, which leaves them vulnerable to human error and bias – a problem which well-trained AI can eliminate, but you may buy into model error and data bias which actuaries can help to avoid. Many AI implementations that can improve risk evaluation are already in use, from handling simple claims-related tasks and better detection of fraud, to preventive initiatives facilitated by smartwatches and other devices.

We can already see that AI developments are likely to have a wide impact, and that they have the potential to send shockwaves into unexpected places.

As this change is already in progress, a full understanding is key to preventing negative outcomes and optimising the benefits. With this in mind, The Actuarial Association of Europe (AAE) has published a discussion paper titled “AI and the opportunities and challenges it presents to insurability”. This paper examines the prerequisites that make a risk insurable, as well as the mitigating factors which can impair insurability, and also takes a deep dive into the implications of AI on all of these.

This article first appeared in the AAE Blogs on February 20, 2023: https://actuary.eu/ai-ok-blog/

You are here

AI: OK?