Public OpenAI services in consumer-facing applications like ChatGPT (which uses input text to create stories or outlines) and DALL-E2 (which uses input text to create images) have gained traction, but  may not be fully compliant for enterprise business usage.

"IntellectAI's genAI architecture is purpose-built for insurance, with protected interaction to Large Language Models (LLMs) that offer industrial-grade security to safeguard your data against external access and misuse" says  Sandeep Haridas, business head of Insurance at IntellectAI. "It is mind-blowing technology," says Haridas, "and businesses in wealth management, capital markets, banking and others — including insurance — are just beginning to explore how it can improve their business outcomes."

Generative AI's untapped insurance potential

GenAI has practical applications in many insurance use cases with its power of multi-modal analysis, and evidence based reasoning and decisioning. Haridas notes that underwriters can leverage this power in new ways, such as generative AI recommended underwriter file content based on submission data and additional insights, and performing Primary Policy Language analysis to determine the quality of underlying coverage. Wholesalers can apply genAI to analytics to get generative business intelligence, which uses natural-language queries to instantly surface matches to the right capacity providers.

"Perhaps contrary to popular thought, adding genAI-based capabilities won't break the bank," said Haridas, who notes that practitioners are just scratching the surface on using the technology. "In fact, genAI can be added into legacy infrastructure pretty easily, depending on how API-ready the legacy environment is."

While adding the technology to infrastructure may be straightforward, attempting to DIY gets a bit more complex. Haridas points to two main reasons:

1. GenAI is incredibly powerful, but there needs to be a clear vision of what outcome the business wants to accomplish. Understanding the positive and negative implications of genAI is crucial prior to initiating such programs.
2. GenAI is prone to hallucinations, which generally result from implicit problems like biases and lack of transparency in the training data. This risk has increased the need for traceability and explainability in AI derivations, and the need for human review and guidance.

Delivering a multimodal model for the insurance industry

Another barrier to genAI lies in the model itself. As opposed to text-based models, which uses a single-modal model of synthesizing text to return text to the user, Haridas points to a use case such as loss runs that require an AI model to be able to "see and read" documents — i.e. synthesize text and layout (images) — and extract the information. This multi-modal requirement would push the limitations of a vendor that is simply connecting to a single-modal (text) model. "The best partner is one that offers the right infrastructure, talent, approach, scale and models refined for insurance use cases that are most important to all the stakeholders," Haridas says.

It's well-understood that the insurance industry is not an early adopter when it comes to new technology. But given all the benefits and use cases for genAI, as well as the ease of adding genAI capabilities to existing infrastructure, those in the insurance industry can't afford to wait on such a promising technology.