For today’s insurers, the successful implementation of generative AI technologies will require balanced experimentation, strategic partnerships and responsible adoption.
In a recent episode of the Insurance Unplugged podcast, Robert Pick, Executive Vice President & Chief Information Officer at Tokio Marine North America Services and Group Deputy CITO for Tokio Marine Group globally, joined host Lisa Wardlaw to talk about how insurers can navigate the opportunities and challenges of AI.
In their discussion, Pick offers a seasoned IT perspective on the opportunities of generative AI. At Tokio Marine North America, he oversees the full technology suite, from enterprise architecture and applications, to analytics, infrastructure and security.
In this except from his conversation with host Lisa Wardlaw, we highlight where Pick sees AI making a tangible impact in insurance today, the red flags insurers must be able to navigate and why companies should be experimenting safely and leveraging their partnerships for expertise.
Practical Insurance Use Cases for GenAI
Pick puts the use cases for AI into two major categories: technical and business.
On the technical side, the starting point is helping insurance technology teams take advantage of generative AI (GenAI) to improve how they work. Pick says that while we cannot expect GenAI to complete tasks, it provides tremendous value in getting you closer to the finish line.
He cites a specific use case around taking stored procedures and converting them to Java. “As a company, we are a universe of stored procedures, and being able to do this is a game changer for us. I don’t mean that the generative AI clicks in and it’s done; it gets us further along than we would’ve otherwise been able to get…60{7df079fc2838faf5776787b4855cb970fdd91ea41b0d21e47918e41b3570aafe} or 70{7df079fc2838faf5776787b4855cb970fdd91ea41b0d21e47918e41b3570aafe} across a thousand stored procedures that might otherwise take two to three or four weeks each. So, it’s a game changer.”
Data lineage is another use case. While GenAI might hallucinate, it never forgets. You can input your reports and code into a safe container and then ask it “tell me where this element is used and how it’s used. Tell me it’s original and where it’s deposited.”
And while GenAI doesn’t understand language, Pick says, “the pattern detection behind it allows it to create insights that are akin to understanding language. Hugely helpful.”
On the business side, Pick cites the advantage for search and how it takes us from search to “answer.” “I’m not going to go search. I’m not going to get a list of links or list of maybes, I’m going to get an answer. Now that answer may or may not be fully correct, but I’m going to get that.” This is extremely useful in aggregating underwriting guidelines and manuals or even in terms of helping train new underwriters or those who are just learning about a particular line of business.
Another area of utility is being able to make sure you’re using the proper language in a particular disclaimer or exclusion for a certain city or state—things that would be difficult for a human to remember. He says: “It’s not intelligence, but it’s similar to intelligence, where applying that to our insurance business capability needs and business use cases could potentially be a game changer.”
Navigating the Red Flags
When it comes to GenAI, the industry has reacted differently compared to other tech innovations. Rather than playing wait and see, insurers have leaned in and said, “Whoa, how can I get involved in this? Even if it was just to observe and not actually experiment and do POCs, everybody said, this seems different and very interesting. I hadn’t seen that before.”
And the claims organization—which is known for its very clear rules, approaches and processes—are aware of the potential of GenAI and they are curious enough to want to try it out.
Bias and hallucinations are certainly a concern. In addition to the incomplete data and inaccurate data that contributes to bias, Pick highlights another aspect of bias that is specific to the insurance industry: “Bias is based upon incomplete data, inaccurate data, but the third way is when you have data that’s complete and accurate, but it gives you an answer that you don’t want to know. In our industry, once we know something, we can’t not incorporate it. We have to know how to deal with that.”
However, the biggest “brick wall,” according to Pick, is drift. As a regulated industry, insurers have to show their work, and this becomes more complicated with GenAI.
“With classic AI, which was really rules-based, basically if the NL scaled up, you could literally go in and say, this is the algorithm. These were the inputs, this was the model or the rule set as it existed on that date, and you will get the same answer every time, time after time. With generative, you can issue the same prompt even hour by hour and you may get different answers. It all depends on how the LLM is being managed and all the different components in between it. And then when you throw a little RAG in there, if you’re ragging in a different way, you can get a lot of varying answers.”
This is something that the industry must figure out. Pick highlights the responsibility that insurers have to regulators, policy holders and shareholders to “make sure that this stuff is clean, understood, repeatable before we start throwing it in line.”
The Call to Action
Pick’s call to the industry is simple: Don’t freak out and don’t ignore it. “Experiment safely, leverage your partners, whether they’re an SI partner, whether it’s a platform or SaaS partner, your partners probably know more than you do. That’s certainly the case for us. Leverage them, learn your way into it.”
While it’s the industry’s responsibility to be stable, this doesn’t mean that insurers cannot try new things, as long as they are doing so safely.
“In my opinion, if you’re not at least dabbling and you’re not at least conversant, you are going to be left behind. Pick your tech stack of choice. There are a bunch out there. Make it safe, try it out. Let your folks try it out safely.”
Conclusion
In conclusion, Robert Pick’s insights offer a compelling roadmap for insurers looking to navigate the complexities of generative AI. By emphasizing the importance of balanced experimentation, strategic partnerships and responsible adoption, Pick provides a practical framework for leveraging the opportunities presented by AI while mitigating its inherent challenges. His call to action encourages insurers to take pragmatic approach to AI adoption, one that encourages insurers to embrace experimentation, leverage expertise through strategic partnerships and prioritize safety and compliance in their pursuit of innovation.
Listen to this episode of Insurance Unplugged with Robert Pick.
