Artificial intelligence (AI) offers the opportunity for insurers to cover more risks at a reasonable price, but it also threatens insurability, as Esko Kivisaari discusses with Christopher Cundy
AI is already being used by insurers in applications such as fraud detection, claims management and natural language processing (NLP) to help improve process efficiency.
But there is another growing use – in actuarial pricing and risk models – which poses deeper questions for the sector.
AI has the potential to improve the understanding of risks, which could lead insurers down two paths.
One is where the technology helps broaden the addressable market by providing coverage to more kinds of risks and at more reasonable prices.
The other path is where the classification of risk becomes so precise that risk-sharing becomes obsolete, and the information asymmetry between customer and insurer tilts too far in favour of the insurer, resulting in discrimination and exacerbating social exclusion.
These issues and more were the subject of a discussion paper published earlier this year by the Actuarial Association of Europe (AAE), AI and the opportunities and challenges it presents to insurability.
One of the report's authors – Esko Kivisaari, the retired deputy managing director of the Federation of Finnish Financial Services and past chair of the AAE – shared his reflections with InsuranceERM on the topic.
Not a revolution
AI has been the subject of decades of science fiction, but with today's access to massive amounts of data and huge computing power, some of those predictions of its capabilities are becoming reality.
Will it prompt a revolution in the insurance sector?
"I wouldn't call it revolution. I would call it a great aid to do things in a better way," Kivisaari says.
"When I started in insurance, data was expensive, and even when it was available, it took a lot of time and energy to manipulate the data. Now we have cheap data and the speed to manipulate it. Hopefully with AI, we'll have more tools to analyse the data."
But Kivisaari is sceptical AI will supersede the work of actuarial modellers, and users of AI should proceed with caution.
"Data scientists and AI people need to trust actuaries because we have already made the mistakes"
"There are those who say that with AI you don't need models anymore: you just ask the questions and AI will do the analysis. Until proven wrong, I have difficulties in believing that. One must be careful with not mixing correlation with causality, and understanding where your model works and where it doesn't," he says.
"There are lots of dangers there. Sometimes I say that data scientists and AI people need to trust actuaries because we have already made the mistakes and others don't need to repeat them."
Positive aspects of AI
Insurance works best - and is most affordable - when there is a large pool of similar risks.
The AAE's report says AI could help with pooling similar exposures across national borders, for example by using NLP to harmonise and simplify products. As well as having a bigger pool, this approach could help extend coverage to countries that are underinsured.
"AI gives much better tools to understand the risks"
Insurers also require a calculable loss to offer coverage, and AI opens up the possibility of bringing far more and real-time data into risk pricing and modelling. For example, the report says images, text or data from connected devices will make more accurate predictions possible, and enable the expansion of usage-based insurance.
"AI gives much better tools to understand the risks, and insure things that might have been uninsurable in the past," Kivisaari says.
The AAE report says one concrete example is replacing strict exclusions from health insurance due to pre-existing conditions with a more nuanced underwriting that accounts for dietary and sport habits.
"Also, AI brings efficiency, meaning that risks that from an administration point of view were too expensive to insure previously, could be insured," Kivisaari adds.
A more accurate understanding of risk could also allow insurers to better segregate risks, and this is where a sinister side of AI might manifest. If insurers then decide to penalise a "bad risks" pool with higher premiums, this will reduce insurability.
But would the insurance industry really allow risk sharing to become obsolete?
"If insurers only target the most profitable risks and not do anything else, then of course that would not be good for our society," says Kivisaari.
The AAE's report is unequivocal that "in all cases insurers should guarantee that the use of AI generally expands insurability and creates a more scientific foundation for their underwriting" and sets out some principles for responsible use (see box, below).
Given the onus on actuaries to uphold this position, the AAE is planning a follow-up paper addressing the role of the actuary in the development and use of AI.
When using AI, actuaries and insurers need to ensure technologies are used responsibly.
This means that:
- the design of the systems is done in a responsible manner;
- models are thoroughly tested, using advanced and documented standards, in order to avoid all kinds of biases and technical errors;
- exceptional care is taken to make sure the models do not cause harm to vulnerable groups, with extra care in relation to cover that is essential for social inclusion;
- when inclusiveness cannot be achieved with private market solutions, the problem should be flagged to appropriate stakeholders with a view to creating solutions utilising, for example, a public-private partnership; and
- the models need to be made transparent and understandable.
Source: AAE Discussion paper, AI and the opportunities and challenges it presents to insurability