Artificial Intelligence & Machine Learning

Addressing Accuracy Challenges in Gen AI Could Be Complex

Gehrmann of Qatar Insurance Group Says AI Hallucinations Cannot Reverse the AI Tide
Addressing Accuracy Challenges in Gen AI Could Be Complex
Lars Gehrmann, group CDO, Qatar Insurance Group

Generative AI has made remarkable progress in recent years. But the enthusiasm for gen AI adoption and the proliferation of its use cases across industries are countered by challenges concerning accuracy and reliability.

Gen AI's ability to identify patterns in its training data allows it to produce output that is convincing, even if it is not entirely accurate.

According to Lars Gehrmann, group CDO of Qatar Insurance Group, AI often communicates in a convincing manner, akin to a talk show host, but does not always provide accurate information. This can be misleading. In the insurance sector, where precision and accuracy are crucial, this can pose a significant risk.

Gehrmann also highlighted that beyond accuracy, variability in language and style, even with the same underlying data and facts, poses challenges for the users. "This particularly causes problems while communicating with our clients across different channels. We aim for a more standardized approach," he said.

Qatar Insurance Group has multiple gen AI pilots underway, ranging from marketing and employee productivity to call center management and insurance claims processing. AI's tendency to hallucinate cannot reverse the "AI tide" as businesses bet big on integrating LLMs into numerous business processes.

The limitations of the technology have triggered a debate in the industry on whether enterprises need a blanket of 100% accuracy in everything. From an insurance claims perspective, Gehrmann emphasized that absolute accuracy may not be necessary in all cases. He suggested that in many instances, reaching around 85% accuracy could be sufficient for straightforward claims.

Evaluating car damage claims serves as a prime example. When assessing a damaged car's picture, the evaluator cannot see all the details. They can identify potential damage to specific components that may need replacement, but this is not fully verified until the car is taken to the workshop. Introducing AI into this process allows for much faster evaluations.

"This is where I see a manifold impact," Gehrmann said. "AI can help humans work 10 times faster, allowing them to focus on more complex products and processes. What was once complicated is no longer so with generative AI."

But human intervention is still required due to regulatory requirements or the nuanced nature of certain insurance assessments.

Current gen AI tools are still in their developmental stages. By assessing accuracy in the context of specific business needs and objectives, organizations can optimize the return on investment in AI, ensure alignment with regulatory requirements, and maintain the integrity of their processes.


About the Author

Shipra Malhotra

Shipra Malhotra

Managing Editor, ISMG

Malhotra has more than two decades of experience in technology journalism and public relations. She writes about enterprise technology and security-related issues and has worked at Biztech2.com, Dataquest and The Indian Express.




Around the Network

Our website uses cookies. Cookies enable us to provide the best experience possible and help us understand how visitors use our website. By browsing cio.inc, you agree to our use of cookies.