P&C industry must consider ramifications of learning technologies before adoption

The lure of learning technologies must be balanced against practical, implementation and ethical issues before these technologies can be fully exploited by the property and casualty industry, David Crozier, president and CEO of Everest Insurance Company of Canada, told attendees of the 3rd annual Insurance-Canada.ca executive forum in downtown Toronto Monday.

Learning technologies – such as artificial intelligence, intelligent machines and cognitive computing – “are, in some form, here to stay… and we have to deal with it,” Crozier said during a forum panel discussion.

He cited a recent Deloitte report out of the United Kingdom that found, among other things, 56% of the processes in U.K. financial institutions could be handled by learning technology today. “That means that 56% of the functions in the Canadian financial services could be handled today. That has huge ramifications in and of itself,” Crozier told attendees.

“The lure to me of having risk thought about, addressed, being serviced 24 hours a day, everyday, without the attendant HR consequences that could come with having people do that is almost too good to be,” he said. “It’s irresistible in theory, but in reality, there remain practicalities, there remain establishment and implementation issues,” and even ethical challenges “that we have to overcome before we can fully exploit these technologies.”

Crozier pointed out, for example, that a machine may look at something so logically, it might unintentionally redline or discriminate. Does the machine have, or can it have, the capability to determine, “I can’t go with what’s logical; I’ve got to go with what’s real,” he said.

Anwar Haneef, financial services partner for IBM Watson, a cognitive technology that processes information more like a human than a computer, noted during the discussion that the training time for Watson to perform certain functions depends on the complexity. While it may take as little as a few months to train Watson to perform a customer service function – because there are finite boundaries on the domain of knowledge and training – it will take longer for more complex functions, Haneef said.

Crozier noted that in a difficult market, the p&c industry must always be conscious of cost constraints. Is a new idea the latest bright, shiny object or is it, truly, the thing that will transform the industry?

“We have test this,” he said of learning technologies. “Often our response, particularly in the insurance industry, to new technology, products or processes can run the extremes. We may embrace the new idea, not even think about all the potential ramifications of it, or at the other extreme, we might bury our head in the sand and hope it just goes away.”

With respect to the cost of implementing learning technologies, “my comment is it will never be cheaper. It’s an emerging technology,” said Mike Fitzgerald, senior analyst, insurance for Celent.

“Insurance technology is no longer a game of blackjack, it’s not 21. It’s a roulette wheel and you have to make many bets,” Fitzgerald suggested to attendees. “Our whole way of thinking about how approach these projects has to change from blackjack to roulette,” he recommended.

“From a risk analysis, pricing, product design, claims handling to fraud detection, there’s a growing place for technology and for learning technology,” Crozier emphasized. That said, there is also a place for people and the expertise they bring to the table, including in situations where automated or intelligent processes require more or less intervention.

It may be that the learning technology is used as a tool – one that is able to collect and categorize information – for those within the industry to help them better serve customers and clients. “Trying to help the underwriters or lawyers, etc. be more productive is a really great way to think about this,” Celent said.

Haneef noted that a cognitive computing system learns from an experience and then sees “something similar or identical,” it will then know how to classify it. In addition, depending on the system, it can indicate the level of confidence for the response it has provided.

There are systems that “can learn an almost infinite amount of information, so anything it can take in, it can then make a decision on,” Crozier pointed out.

But it is critically important that those in the insurance marketplace share information more than they currently are. If only parts of the story can be compared, “it may not come out with the right answer. So we have to share in the interests of the consumer,” he said. Once the information is “anonomized,” Crozier told attendees, “it just gets used.”

More coverage of the Insurance-Canada.ca Executive Forum

Autonomous vehicle regulation in Ontario will ‘come to fruition,’ MTO official says

Some businesses still do not believe they will be/could be a cyber target: panel