You can easily understand how much the world has changed since the turn of the millennium, because you can hardly even imagine how you were able to live without Google, Amazon, Facebook and Smartphones, just 17 years ago. Almost everyone uses search engines, online trading and social media platforms every day. And almost everyone has these services permanently within reach on their Smartphone.
With these services, artificial intelligence (AI) has entered our everyday life in a big way. We have long since become accustomed to the highly complex, continuously learning algorithms in phrase searches, phrase additions and context-related suggestions. Our Smartphones effortlessly recognize people in photos and understand the words we speak. Texts are translated into any language, as if by magic. All these services are AI products now in everyday use, which at first astonished us as consumers, but which are increasingly becoming the norm and are now an integral part of our lives. AI is therefore clearly a sustainable trend.
But how do insurers use the same modern AI that Apple, Amazon, Facebook and Google have used to revolutionize the consumer market?
Artificial intelligence and insurers – a current overview
One thing you should know in advance: insurers have been using AI methods for a long time. For decades, OCR systems have been using optical character recognition to read addresses, barcodes, insurance numbers and other information from incoming mail. Expert systems support the classification of documents, the checking of insurance payments, and help with decisions for marketing campaigns. However, most of these systems are currently rule-based, i.e. they obey firmly configured rules. Self-learning systems that make decisions based on training data are not very widely used at this time. So, why is that?
Certainly, one of the reasons is that insurers attach great importance to the transparency of decisions in business processes. So, insurers have certain reservations about a "black box" learning machine, whose decisions cannot be understood by the person in charge and cannot be communicated to third parties. In addition, it is often difficult and costly to provide the high-quality and correctly classified data needed to train the learning machine so that it can generalize from examples and make "good" decisions in the future. Another problem is the lack of knowledge about how to select and properly configure suitable machine learning based procedures. Finally, there is the need for a thorough evaluation or comparison of the results of the learning machine. Often, there are not enough key figures available for this purpose.
However, these reservations seem to be melting away. Many insurers are currently using modern machine-learning-supported AI, fueled by the future Industry 4.0 project and the vast developments in this environment in recent years. People are increasingly aware of the potential that this new artificial intelligence has to offer. Success stories in the media, such as the fully automated motor vehicle expert [1], improved customer service through AI [2] or how machines are able to detect insurance fraud [3] are making people sit up and take notice. We can see the changes in the market and society, the massive innovative power, the many research groups, the start-ups that are developing self-learning systems and are producing a multitude of new ideas. An increasingly greater amount of digital and semantically annotated information is becoming available, which can be used intelligently to make automated and autonomous decisions. It also clear that the use of self-learning systems does not necessarily mean that the system has to be a "black box." They can also be used to optimize rule-based systems or to support them in the form of hybrid procedures.