Technology advancements and digitalisation mean actuaries can better price risks, but how can they ensure fairness? Are there ethical risks that might emerge from the use of external data and machine learning analytics?

Data used to be scarce and expensive, according to Esko Kivisaari, past-chair  of the Actuarial Association of Europe. Traditionally, actuaries have used simple proxies such as gender, age and postcode to differentiate between risks. However, developments in digitalisation mean that huge amounts of data are available now, and the costs of processing have gone down.

“There are tools to manipulate [data], we can draw much insights from better use of data,” he said during the European Actuarial Day 2023 webinar, organised by the AAE and the European Actuarial Academy.

Increased availability of data also raises questions about whether people are treated fairly. 

Kivisaari pointed to actuarial fairness - a concept in the insurance industry that aims to establish fair premiums based on policyholders’ risk exposure.

“What do we mean by fairness? We mean fairness by how clients are treated, but also how the existing pool of insureds are treated, not to compromise the interests of either party here,” he said. 

“With data, we can differentiate between risks, but how do we avoid discrimination when we do differentiation?”

The AAE has developed European Standards of Actuarial Practice. ESAP 1 general actuarial practice outlines how actuaries should adhere to professional standards. For instance, in the model governance session of ESAP 1, the actuary involved in using models should “be satisfied that the model risks have been identified, assessed, and that there are appropriate actions to mitigate these risks, such as adequate model validation, documentation, and process controls”.

However, Kivisaari observed that there have been discussions on whether the standard is up to date with the developments in data science, and queried how actuaries could price the risks correctly without causing any undue harm to the society.

He said applying the concept of actuarial fairness could lead to poverty premiums or to ethnicity premiums in certain cases.

“But that has of course happened previously already. When using, for example the zip code, it has resulted in some areas consisting of vulnerable minorities being charged higher premiums,” he said. 

“We must take care that when using the new technologies, we don't cause additional harm.”

The EU General Data Protection Regulation is currently in place, but the upcoming AI Act is likely to restrict insurers from using automated systems to assess customers’ risks or insurance needs without detailed records. 

Kivisaari stressed the need for transparency in the actuarial profession, because Article 52 of the AI Act emphasises transparency obligations for ‘high-risk’ AI systems, including providing clear and accessible information to users. 

“We have begun to clear things with AI, but we must remember the dangers,” he concluded. 

Critics have voiced concerns over the proposed regulation as some use cases of the insurance industry will be on the list of ‘high risk’, leading to stricter rules in the future.  

Petra Hielkema, chair of the European Insurance and Occupational Pensions Authority, said one of the earliest drafts of the regulation revealed insurance in general as a high risk, “whether you were talking to a chatbox or claims handling... I think that was a concern because there would have been too much of an impact”. 

At a panel discussion at Insurance Europe’s annual conference earlier this month, she said the Act has since improved with a narrowed scope but argued the sector was already highly regulated, with Solvency II in areas such as data quality and model validation. She said Eiopa would continue to work with stakeholders to ensure that the sector benefits from technological advances.