Generative AI: Recent developments and insurance use cases

08 August 2023

Bence Lukacs, a product expert at the digital innovation partner Supercharge, assesses the potential of generative AI and considers what initiatives such as ChatGPT could hold for the future of insurance.

Few technology developments in recent years have triggered more debate than the growth of artificial intelligence (AI). Seldom far from the headlines this year, AI is literally taking on a life of its own, and with the much-publicised launch of ChatGPT, we are facing big questions about the rewards – and the risks – of a technology that could change the world.

“We’ve reached an inflection point,” says Bence Lukacs. “With the creation of Large Language Models (LLMs) such as ChatGPT, we can build so-called ‘killer technology’ applications across all industries, not least insurance. The term refers to a radical innovation that will quickly obliterate the usage of the techniques or processes that came before it. Until now, AI has played a limited role in the insurance industry, mainly in specific and narrow applications, so it has not been transformative. However, using natural language processing, AI technology could open possibilities for a broad range of new use cases.”

The power of language

While ChatGPT is the name on everyone’s lips at the moment, it’s more interesting to look at the technology behind it and how language models are shaping a wide range of advanced AI solutions. As Lukacs explains, ChatGPT is a member of the generative pre-trained transformer family of language models (hence the GPT abbreviation).

“Generative AI includes AI models that can create new text, images, video, audio, code or other types of data, based on the material on which they were trained,” says Lukacs. “ChatGPT is an example of LLMs that take a piece of text as a prompt or instruction and then generate a likely answer that fits the context. With narrow AI, you can only perform limited, specific tasks, whereas language models can perform complex tasks across a much broader domain.”

Pioneered by a growing number of Big Tech companies, LLMs use a deep-learning AI algorithm that processes massive data sets to generate new content.

Trained intelligence

As Lukacs explains, the intelligence of generative AI comes from its extensive initial training on a vast amount of data − the ‘pre-trained’ element of the GPT abbreviation: “We talk about natural language processing,” he says, “because that’s what these language models do. They process natural language text in any language that they are trained on. They follow an instruction, a prompt, from the user; it could be a typed question or an instruction, from which the model generates an answer based on the vast amount of relevant text on which it was trained.”

As well as drawing on vast amounts of information from the public domain, such as internet content, Lukacs adds, you can augment the knowledge of the models with your own data and information. By way of example, an insurance company could build a chatbot that helps answer customers’ questions about a policy.

“Integration with your own data sources and systems will be a key strength,” states Lukacs. “If you connect a language model to your own systems and data sources with an API, you can harness the amazing capabilities of the AI model for your own specific use cases.”

Insurance use cases

Lukacs argues that generative AI is unlikely to eliminate the need for direct human involvement in critical business processes anytime soon. But there are many ways in which it can supplement and improve traditional insurance practices. Looking at the current interest in the technology, Lukacs sees several clear benefits of the underlying language model family: “At the basic level, you could create highly flexible chatbots for everyday routine insurance tasks,” he says. “For example, summarising insurance documents for readability, and generating replies to customer queries based on your own knowledge bases. With simple integration via APIs, you could analyse and triage claims, and employ a GPT-enabled chatbot assistant. Finally, using advanced API integration with core systems, you could automate the claims and settlement process, and streamline quotes.”

Understanding the risks and limitations

Despite the many positives of AI, we are increasingly hearing about the dangers if it evolves too quickly and without the necessary controls. For insurers, there are risks in over-reliance on non-human interpretations and decision-making: “We need to be conscious of errors and biases,” warns Lukacs. ‘’It’s what we call hallucination or confabulation. This is when the model generates fake answers because it doesn’t have a deep semantic understanding and instead uses probabilistic calculations to generate output. This can sometimes go astray, which is why human supervision and assessment is still important.”

Building for the future

According to Lukacs, we should consider the new generation of AI models as foundations that need to be continually refined and developed to make them fit for purpose: “If you take the GPT model and apply domain-specific training, you can increase its performance and accuracy on very specific tasks, like detecting fraud in an insurance claim or answering a query about an insurance policy, he says. “With additional customisation and fine-tuning, you can increase the reliability and performance of foundation models for your own use cases. That’s the real power of these models.” 

About Supercharge

Supercharge is an innovation partner that helps clients realise the benefits of digital technology. It works worldwide across many industries, including insurance, and designs, engineers and scales digital products to make businesses fit for the future.

Blog posts

Clearspeed’s intelligent voice-based screening technique aims to build trust, accelerate claims processing and minimise fraud. Alex Martin, Clearspeed CEO, explains the value of voice analytics.
Customers are driving insurers to engage more with AI-powered automated messaging technology like chatbots. Johannes Seebacher, co-founder of software provider riskine, describes their benefits and how they can even be used to offer risk analysis for customers.

User questions

Answered questions

Unanswered questions

Please note that submitted questions are public.

Views: 7061

Downloads: 0

Ratings by number of stars:
0 %
0 %
0 %
0 %
0 %

Page is favored by 0 user.

Contact inquiries: 0

Rate this page!