How emerging AI could impact the insurance sector

Authored by Travelers

Among the many new technologies to enter the public domain in recent years, generative artificial intelligence (AI) feels especially futuristic. Unlike traditional AI systems that recognise patterns and make predictions – such as what programmes you might enjoy watching on Netflix, for example – generative AI creates new content in the form of audio, video and text. It’s helping data scientists across sectors envision many new possibilities for improving how we do work.

In a recent webcast, Mano Mannoochahr and Girish Modgil, two data science leaders from Travelers, addressed the potential that AI tools offer insurers, brokers and insureds alike – as well as the risks that users of this technology must bear in mind.

First, here’s a little about how generative AI works.

Recent examples of the technology include ChatGPT, the text-based generative AI launched late last year, and the newly launched GPT-4, which is a more advanced version of its predecessor. ChatGPT has garnered breathless news headlines for its ability to write poems in the style of Shakespeare, summarise books, write code and even pass medical licensing exams. That’s because ChatGPT has been trained on a majority of the internet (about 500,000 books), 12 programming languages and 499 billion words. This allows ChatGPT to deliver curated responses while sounding confident and conversational. GPT-4, which is available to premium users of ChatGPT and has more recently been powering Microsoft’s Bing chat functionality, is reported to be able to answer difficult questions with greater accuracy and provide fewer false responses than ChatGPT. Its developers also built a layer of ethics into it, so it will avoid answering questions that it deems harmful.

Using this technology works much like making a query on Google, but it generates different outcomes. Since ChatGPT has been more widely available at the time of this writing, let’s compare its outcomes to those from Google: A Google search will deliver consistent answers and a wide variety of results with repeated uses – and the user must then judge the veracity and utility of those results. However, similar queries to ChatGPT could yield a different answer each time. Sometimes the answers will improve with time and sometimes ChatGPT will just make up an answer – even if the user is searching for a fact as clear-cut as a citation from a research paper. ChatGPT provides an answer that is superficially good and gives the impression of confidence, but the technology can’t actually reason. We have to consider its answers with a healthy dose of scepticism.

Risk and reward

Still, the technology has important potential in augmenting the work of humans. This is where Mano Mannoochahr and Girish Modgil see it being especially useful to insurers in the future.

Of course, insurers already rely on data science capabilities for risk pricing and product segmentation. But generative AI can help insurers rethink all aspects of the business. Consider claims. Generative AI could enhance proprietary claim damage models so that following a CAT event, insurers could run models to assess potential damage to properties, then deploy adjusters and claims handlers to prepare for an onslaught of calls in a particular area. After a fire or flood, we might be able to gather thousands of images of the disaster and start a claim for a customer before they have even been able to return to their business and survey the damage.

AI-based tools have the potential to support disciplined underwriting as well. They could help us chase data about a property without taking up the time of a broker or insured – and just provide faster turnaround times in general. Imagine an underwriter or claims manager coming to work in the morning to find that AI-based tools have generated reams of data overnight that they can now use to provide meaningful responses to queries from brokers.

More generally, there are applications for AI-based tools to support the smooth administrative function of a business. These tools can onboard new employees who need basic training to get up to speed about the insurance industry. They can provide documents that are a good starting point for job descriptions, customer welcome letters or business plans.

But to be effective, these tools need the oversight and direction of humans. Despite what science fiction novels might make us believe about technology like this, AI is not here to replace us but to help us make better decisions and provide a higher level of responsiveness and service.

AI tools on their own can’t deliver this safely. In the case of ChatGPT, it’s currently worth trying as a fun experience but isn’t something to rely on for important information. For instance, asking certain questions of the tool can help to train it on your business and potentially allow a competitor to generate knowledge specific to your company. Even if you ask the technology to generate code, a person must check the code’s accuracy. Based on early reports of GPT-4, the same is true of this advanced version of generative AI. The technology must augment the human.

The combination is powerful. There is so much opportunity ahead of us to use AI to collect more information more quickly, then use it to improve the lives of insureds. It will help us do an even better job of protecting what matters to them.

ChatGPT is available for free to the public via the website OpenAI.com, while GPT-4 is currently available to ChatGPT Plus users. OpenAI.com also developed the image-based generative AI models DALL-E and DALL-E 2. Other tech companies are developing their own generative AI tools as well, such as Bard (from Google) and LLaMA (from Meta).

Jon Davies is Vice President International Management Liability at Travelers Europe.

The information provided in this document is for general information purposes only. It does not constitute legal or professional advice nor a recommendation to any individual or business of any product or service.