OK, OK, Yet Another Item on AI

By: Dr. Gary Anderberg

We refer to the time over 11,000 years ago as an Ice Age because there was ice everywhere. Welcome to the AI Age. A recent item in ITL looked at how AI is shaping developments across risk management. Yup, it's everywhere. The article starts by noting that the insurance industry has been relatively slow to adopt AI compared to many other industries.

But that's insurance for you. Wrapped as we are in so many rules and regulations, multi-year contracts/policies, actuarial tables, and so forth, we have to proceed carefully. Our industry is largely built on precedents and data troves of long experience. AI tries to upset all our apple carts. And, sometimes, that's just what we need.

Well, it's getting there. This compact, nicely written article looks at four areas where AI is beginning to show results.

• Claims processing: AI is proving to be the adjuster's ideal little helper by speeding along all sorts of secondary, previously manual activities, especially those involving correspondence and document management/lookup.
• Fraud detection: It's still in the early days in this area, but AI is opening up new ways to review claims and spot suspicious events (or non-events*).
• Reserving and underwriting: We probably have the best running start here because first-generation machine learning AI moved reserving and underwriting way ahead of old-fashioned rules engines a decade ago, but new AI developments allow us to add important nuance to get both reserving and underwriting right.
• Knowledge transfer: This is the biggest area where AI can bring huge benefits. The last of the Boomer generation of adjusters — those folks who have seen everything and are afraid of nothing —is retiring. AI is just in time to help us capture their knowledge and make it optimally available to our newer adjusters. This area is where AI will earn its accolades. Just watch.

And the beat goes on. Shiny new OpenAI o1 is undergoing trials now. The improved "reasoning" abilities of this release are being tested by Oscar Health to aid in solving a number of problems concerning reimbursement methodology for complex claims in group health administration. These issues aren't significantly different from problems we encounter in complex medical bill repricing scenarios in workers' comp. How about that.

The AI revolution is well into its hype followed by reconsideration/disappointment cycle right now. We start by expecting too much and then lapse into relative depression when we find our new toy has limitations. But AI is still being invented. OpenAI o1 is a good example. Expect more items about AI in these pages. We are just starting on a very long and fascinating trip, folks.

*One reason finding fraud can be difficult is that sometimes the most suspicious matter in the file is what didn't happen, like the Sherlock Holmes case of the dog that didn't bark.