Frankfurt The use of artificial intelligence (AI) is playing an increasingly important role in the insurance industry. Especially in claims processing, AI-based systems help to simplify and significantly accelerate settlement. However, Stephen Voss, co-founder of the insurance start-up Neodigital, warns that criminals will also increasingly use artificial intelligence: “The new possibilities surrounding ChatGPT will also lead to an increase in fraud.”
In contrast to previous AI-based chat programs, the new chatbot from the US start-up OpenAI writes texts that could come from people. Even if the generated texts sometimes contain incorrect information, it is clear which developments are possible in the future.
According to Voss, fraudsters will also take advantage of this. Criminals could now describe a claim more and more easily and have a suitable photo selected: “The better the artificial intelligence gets here, the more difficult it will be for claims adjusters at insurance companies to recognize these cases as fraud.” There are still only a few AI-generated fraud attempts submitted. However, Voss expects the number of these cases to rise quickly.
Insurance fraud is not a new phenomenon, even if there are hardly any concrete figures. In a survey commissioned by the insurance association GDV in 2020, 5.6 percent of the participants stated that they had reported incorrect damage themselves.
Another 3.8 percent of those surveyed said they had received a request to report incorrect damage. Most frauds occur in personal liability and household insurance.
AI systems recognize dubious damage patterns
Conversely, the survey also shows that over 90 percent of insurance customers are honest. Voss from Neodigital also emphasizes this: “The vast majority of the claims submitted are of course absolutely justified.”
The chatbot ChatGPT from the company OpenAI can write damage reports.
(Photo: Reuters)
Neodigital is one of seven insurtechs that the financial regulator Bafin has granted an insurance license for since 2017. The digital insurer sells various property and accident policies and cooperates with Huk-Coburg and HDI Versicherung, which are also involved in the start-up. The investors also include a fund run by financial entrepreneur Carsten Maschmeyer.
“But we also have to protect our policyholders from the honest community being harmed by criminal attempts at fraud,” Voss continues. Ultimately, higher claims payments also lead to higher insurance premiums.
Insurers, on the other hand, also see great opportunities for artificial intelligence in preventing fraud. AI systems can detect dubious damage patterns or discrepancies in submitted photos. Allianz insurance, for example, announced last year that AI 2021 had prevented payments in the three-digit million range.
>> Read also: Nobel laureate on AI programs – “There will be people whose qualifications are no longer in demand”
Voss explains that the metadata of photos is evaluated, among other things. They show when a picture was taken. If it occurred before the damage occurred, fraud is quickly apparent.
AI detects image manipulation
Insurers also use reverse searches on the Internet, which consumers can use on their smartphones. According to Voss, for example, the photo of a destroyed ceramic hob, which can be found on a web portal for cooking and baking, has often been submitted, as has the picture of broken glasses by soccer coach Jürgen Klopp. Insured persons would no longer be able to get away with this today.
Images can now be manipulated very well. According to Voss, glass damage can be professionally repaired afterwards. In order to reliably detect such attempts at fraud, the AI must be constantly trained.
>> Read also: AI stocks – With these four ETFs, you can bring the hype into your portfolio
Voss has so far advised against the fully automated processing of claims – for example simple claims in the motor vehicle sector – even if it is now technically possible. In certain cases, Neodigital continues to rely on a final check by a human clerk. In this way, the insurtech wants to prevent fraudsters from regularly submitting smaller claims written by an AI that remain below the threshold of perception.
He sees a challenge for insurers in the fact that employees are often well trained in identifying cases of fraud written by people. “A lot can be read between the lines here,” says Voss. For example, people often become suspicious when damage reports are formulated too perfectly.
However, an AI can adapt to this quickly and deliberately incorporate errors so that the fraud is no longer noticed. The Neodigital board is convinced that insurers are not yet well enough prepared for this development.
More: An overview of features and disadvantages of OpenAI’s AI