The advent of ChatGPT has highlighted the ability of generative AI to push out new content. Insurers are already experimenting with the new technology to find out how it can be used. Asia Insurance Review spoke to some experts about the new developments.
There has been much hype and caution behind the power of generative AI to create content, codes, artwork and even poetry. However, the new technology also has potential to help insurance professionals in their work.
“Insurers should certainly be cautious of any new technology such as ChatGPT,” said RGA data science vice president Jeff Heaton speaking to
Asia Insurance Review.
“I believe that this new technology also brings great potential. It has the capability to act in an assistant role to a variety of insurance professionals. However, it lacks reasoning and common sense. Because of these limitations, it can be confidently incorrect and any results provided must be checked by a knowledgeable subject matter expert.
“In my own experience, I have used ChatGPT to assist in writing routine portions of data science analysis code. Generally, it is correct but does require checking. It currently lacks the ability to incorporate large amounts of internal proprietary company data and program code, which is a considerable limitation,” he said.
Mr Heaton said that RGA is currently in the early stages of evaluating ChatGPT primarily to assist insurance professionals with routine tasks.
“I have not seen a public insurance ChatGPT-related project; however, Microsoft is working on a variant called BioGPT, which contains more training around biology than the general ChatGPT,” he said.
Speaking to
Asia Insurance Review, Microsoft Singapore senior solution specialist (cloud) Yassine Serhane said, “We’ve seen extensive interest in Azure Open AI Services including ChatGPT … For example, some of our customers in the insurance industry are testing out generative AI to strengthen support for customer help desks. They are also turning to AI to make sense of unstructured data from documents to assist in claims processing.”
Potential of generative AI in insurance
Dr Serhane shared some possibilities that generative AI could have in the insurance industry:
- Customer service: Automating customer service tasks, such as answering frequently asked questions, helping customers with policy information, and providing support for filing claims. This helps insurers to minimise repetitive work and focus on business and customer priorities.
- Policy recommendations: AI can be used to provide customers with personalised policy recommendations to help them make better informed decisions based on their individual needs, risk profiles and critical life events.
- Claims processing: AI can simplify the claims process by automating data analysis and providing personalised recommendations on how customers can handle their claims, while also helping to prevent fraud through claims analysis.
- Risk assessment: The use of AI in analysing customer data during risk assessments can provide accurate recommendations on how to best manage risk.
- Knowledge mining: AI can help employees learn more about their customers through enhanced search capabilities into policies, contracts and other complex insurance related documents which can in turn, streamline business processes and improve customer experiences.
“Generative AI has the potential to revolutionize how insurers interact with customers by utilising its capacity to comprehend context, produce natural language responses, ask clarifying questions, handle multiple tasks and learn from feedback. This will enable insurers to provide seamless and individualised experiences for changing customer and industry requirements,” said Dr Serhane.
“We can leverage AI to understand the underlying patterns of fraud, past cases and ensure that the knowledge worker is better informed in making the decisions he or she is making,” said IBM Technology ASEANZK CTO Kalyan Madala speaking to
Asia Insurance Review.
He said IBM had developed AI systems for agents and insurance customers to reduce the number of customer service calls. However, insurers should not overlook other aspects of AI which have been instrumental in claims processing and fraud detection.
“We’ve created a conversational interface for people to submit claims. Where the assistant asks questions about the incident, what happened and what information is needed … if somebody is submitting a claim form, it looks at how the language is worded, how much time was spent by the user to type the claim and how much time was spent read it. A lot of times when it’s a genuine claim, people didn’t spend too much time,” he said.
Challenges for developing generative AI tools
It may take significant time and resources before generative AI tools can be adapted to help an insurer communicate with external stakeholders.
“Chatbots will require considerable internal proprietary information to be effective. This information would have been outside of ChatGPT’s training sources. Additionally, information that ChatGPT may have from the public internet that conflicts with a company’s own information must be removed. Such retraining of large language models requires specific expertise and a considerable amount of expensive cloud computing,” said Mr Heaton.
Insurers will also need to ensure that they keep up with regulatory requirements over the use of AI and ensure that the technology is used ethically.
“As with any technology, there are risks and challenges that need to be considered and addressed in deployment. The same applies to industries with high regulatory and compliance requirements like insurance. When implementing technologies like generative AI, businesses should take extra precautions to ensure the ethical and responsible use of AI. For insurance providers, this would entail reviewing AI-generated policy recommendations before communicating them to customers,” said Dr Serhane.
“We’re very used to implementing deterministic systems and rule-based systems where we know exactly the answer going to be based on the inputs. AI uses probabilistic systems where the response could vary based on the input as well as where the model is in its own learning. We also need to be a lot more cognisant of the risk that exists in this process,” said Mr Madala.
He said insurers will need to build a good framework to address and manage the risks associated with using a generative AI technology. It is also imperative to understand how the AI model works and whether it has generated responses in a fair way regardless of race and religious beliefs.
“It’s an ongoing process to ensure we keep this model at a good performance level ... This is something we must actively monitor and remediate if there’s a problem,” he said. A