Help

BI’s Article search uses Boolean search capabilities. If you are not familiar with these principles, here are some quick tips.

To search specifically for more than one word, put the search term in quotation marks. For example, “workers compensation”. This will limit your search to that combination of words.

To search for a combination of terms, use quotations and the & symbol. For example, “hurricane” & “loss”.

Login Register Subscribe

Machine learning catching on in insurance, but challenges remain

Reprints
machine learning

SAN ANTONIO — Emerging tools such as artificial intelligence and natural language processing are being used in the insurance sector, but costs remain high and there are questions about bias being introduced into machine learning, according to a speaker at the Public Risk Management Association’s annual meeting Monday.

“Everything is smart these days,” said Brian Billings, vice president of predictive analytics in Ballwin, Missouri, for  Midwest Employers Casualty Co., part of W.R Berkley Corp., and such devices as cell phones and televisions now collect data from their users. “All of that technology is being driven by the use of data.”

Machine learning, including artificial intelligence and natural language processing, takes the data being collected and tries to predict some kind of outcome, Mr. Billings said, such as a numerical value or, in the case of the insurance sector, a claims scenario.

With natural language processing, a model is trained to read text, Mr. Billings said. Such technology can take a 40-page discharge summary and extract specific relevant text, such as all doctors’ or lawyers’ names, or medical notes. “It has huge implications in the claims adjusting space.” He noted that his company has such tools in use.

While the technology remains expensive, Mr. Billings is confident the tools will reach a “point of democratization” in their use penetration and ultimately become more affordable. “I think the day is coming where it will be available and not be prohibitively costly to have at your fingertips,” he said.

The introduction of bias in the training of the machines and models remains a concern.

“Bias is an issue, for sure,” Mr. Billings said, adding it is “absolutely” something to watch out for. He remains confident, however, that technology will continue to evolve and be deployed more widely in the insurance sector.

“Technology is not going to slow down,” he said. “The amount of data that we have and generate is not going to decrease. We’re going to see more and more of this.”