The Future of AI Is Now
It’s been an interesting couple of months. In between helping to set up a new IoT insurance center in Munich whilst being caught up in corporate restructuring, I’ve also been catching up on my reading.
There’s a lot in the press about “artificial intelligence” and “cognitive analytics” — in effect the whole issue of machine learning. I’m thinking a lot about how this might not only affect insurance business models but also individual professions within the insurance industry.
Along the way, I also suffered my own insurance claim — I’ll share details in a moment or two — and saw how AI and the “real world of claims” seem to come together.
In my research I came across a very interesting academic paper called “Future Progress in Artificial Intelligence: A Survey of Expert Opinion,” which I would commend to you. The paper, originating at the Future of Humanity Institute of the University of Oxford, points to concerns about the impact of AI on humanity and how it will bring “significant risks.”
If we look at 10 years [from now], then insurance in an AI environment will happen within our immediate working lives.
It makes the point that these risks to humanity have either been ignored in the past or considered to be science fiction.
In my presentations I often refer to the movie “Minority Report,” which is based on the Philip K. Dick science fiction story about predicting thought crime, and I argue that these capabilities are already with us — even if flying through the ether is not yet currently available.
Cognitive analytics are already piloted in health care and wealth management. But the long and short of it is that some major insurers are already thinking hard about cognitive analytics and AI in the context of the Internet of Things.
So when will all this happen in reality?
The experts say that there is a one-in-two chance that high level AI “will be developed around 2040-2050, rising to a nine-in-10 chance by 2075.” They also say incidentally that there is a one-in-three chance that this will be “bad for humanity,” but let’s save the ethical questions for a different day.
These are average figures, but let’s drill down a little more deeply.
Of the group of 123 eminent scientists surveyed, 11 percent say that we will understand the architecture of the brain sufficiently to create machine simulation of human thought within 10 years. Of this, 5 percent suggested that machines will be able to simulate learning and every other aspect of human learning within 10 years.
They also predicted machines will have the specific levels of maturity worthy of a third grade school exam by 2030, the Turing test (in which an observer cannot tell the difference between a conversation with a human or a machine) by 2040, and Nobel-level research by 2045.
So, let’s roll the insurance clock forward either 10 or 30 years depending on how bullish you feel. If we look at 10 years, then insurance in an AI environment will happen within our immediate working lives.
Even if you have retired, you will still be affected as it will impact the way you buy cover and the way your claims are serviced.
If we look at a longer timescale of 30 years, then the grads and interns coming into the industry today will be using this before they finish their careers.
For what it’s worth, I think it will happen sooner rather than later.
But, I promised to tell you about my own claim.
My delayed flight back from Munich to London was much delayed, and I was greeted at about 1 a.m. with a flat battery in my car. I called the auto breakdown insurer who promised a repair guy within 30 minutes who actually turned up in 20 minutes.
Not only did he bring a professional solution in the middle of the night, but he also brought with him a smile, which made all the difference to a tired guy like me. Great service, great attitude.
At the end of the day, isn’t it still going to be about the basics, especially at the claim’s “moment of truth?” I really hope that the AI that we eventually learn to live with in insurance will recognize the importance of service and the very great power of customer advocacy.
FEMA’s Revisions May Result in Less Flood Protection
No one is more impressed than Susan Williams by the shoring up of 50 levees and rehabilitation of flood walls and pumping stations overwhelmed when Katrina struck New Orleans hit in 2005.
Nor is CoreLogic’s content strategist surprised by the result: FEMA map revisions that removed nearly 60,000 homes from the Special Flood Hazard Areas (SFHA) on Sept. 30, exempting their owners from mandatory compliance with NFIP flood insurance requirements.
But Williams’s buoyant mood comes with a caveat.
“If the lender only follows the government guidelines, homeowners would not be required to obtain flood insurance. But just because they’re not in that flood hazard area anymore because of the new mapping doesn’t mean the flood risk has gone away.”
John Elbl, vice president at AIR Worldwide, agreed with that assessment, fearing individuals still at great risk may elect to drop flood coverage from their policies to save money in the short term.
“We’re hopeful private insurers will step in to help fill this protection gap by charging actuarially sound rates and providing policyholders with more comprehensive coverage options.”
The problem may not be private insurers, however, but the banks. A year ago, Elbl noted the opportunities for insurance industry to provide alternative coverage to the NFIP ran into complications, not the least of which was reluctance on the part of banks to accept policy wording that wasn’t identical to the NFIP’s.
That uncertainty, he said at the time, “restricts the ability of the customer to choose private coverage.”
Today, Elbl said insurers considering entering the market to compete with the NFIP are pinning their hopes on expected NFIP rate increases driven by the program’s $24 billion deficit and its inability to offer more than limited coverage for policy holders.
“NFIP policies do not cover basements, do not include payouts based on loss of use of a property, and only pay actual cash value, less depreciation, as opposed to standard HO3 policies which pay out full replacement costs.”
Another bright sign, added Williams, are for properties in areas of greatest risk and who pay the highest rates.
“The cost of flood insurance should go down and hopefully that will in turn make people more interested in getting coverage. That in turn adds to the resilience of the area.”
Overly Optimistic Message
Dean Basse couldn’t disagree more. The general manager at Dan Burghardt Insurance in New Orleans said an overly optimistic message to cash-starved homeowners about the city’s improved levee system will disincentivize them from obtaining flood insurance when they’re no longer required by federal law to do so.
“We can’t give away a policy unless they’re forced to buy it,” says Basse. “They won’t buy flood insurance voluntarily even at the PRP rate.”
Moreover, the infrastructure rehab along Louisiana’s 300-mile coast line, said Basse, may not be the final answer in flood protection.
“They spent a billion dollars to build this giant rock wall and someone forgot that it’s still built on mud. Our mud is not known for being the most solid thing in the world. And it’s sinking.”
It gets worse, Basse added, the next time another major catastrophe hits New Orleans like Katrina, like Betsy in 1965 or the “thousand-year rain” that dropped onto Louisiana this year.
Jackie Noto’s concerns are more for the NFIP itself, which was intended, the model product manager at Risk Management Solutions said, to support the original charter and objectives of the Flood Protection Act.
Unfortunately, said Noto, “the NFIP is 50 years old and its methodologies and approach haven’t aged gracefully.”
What the NFIP must do, she said, is “move away from this `in or out’ mentality. What matters is the depth, severity and frequency of flood risk for our entire area.”
“That’s also been proven in the recent Louisiana flooding as well where buildings that were damaged weren’t considered on plain and still aren’t as recently as FEMA’S map updates. That’s definitely not an adequate approach.”
To those insurers who believe there is no such thing as a bad risk, only a bad price, Noto said their business actually improves the more they are able to differentiate risk.
“When they have information on the level of protection and probability of flood defense failure, that allows them to price risk to reflect the appropriate risk itself.”
Unfortunately, Noto added, “that’s not something the insurance industry is used to.”
Why Marine Underwriters Should Master Modeling
Better understanding risk requires better exposure data and rigorous application of science and engineering. In addition, catastrophe models have grown in sophistication and become widely utilized by property insurers to assess the potential losses after a major event. Location level modeling also plays a role in helping both underwriters and buyers gain a better understanding of their exposure and sense of preparedness for the worst-case scenario. Yet, many underwriters in the marine sector don’t employ effective models.
“To improve underwriting and better serve customers, we have to ask ourselves if the knowledge around location level modeling is where it needs to be in the marine market space. We as an industry have progress to make,” said John Evans, Head of U.S. Marine, Berkshire Hathaway Specialty Insurance.
CAT Modeling Limitations
The primary reason marine underwriters forgo location level models is because marine risk often fluctuates, making it difficult to develop models that most accurately reflect a project or a location’s true exposure.
Take for example builder’s risk, an inland marine static risk whose value changes throughout the life of the project. The value of a building will increase as it nears completion, so its risk profile will evolve as work progresses. In property underwriting, sophisticated models are developed more easily because the values are fixed.
“If you know your building is worth $10 million today, you have a firm baseline to work with,” Evans said. The best way to effectively model builder’s risk, on the other hand, may be to take the worst-case scenario — or when the project is about 99 percent complete and at peak value (although this can overstate the catastrophe exposure early in the project’s lifecycle).
Warehouse storage also poses modeling challenges for similar reasons. For example, the value of stored goods can fluctuate substantially depending on the time of year. Toys and electronics shipped into the U.S. during August and September in preparation for the holiday season, for example, will decrease drastically in value come February and March. So do you model based on the average value or peak value?
“In order to produce useful models of these risks, underwriters need to ask additional questions and gather as much detail about the insured’s location and operations as possible,” Evans said. “That is necessary to determine when exposure is greatest and how large the impact of a catastrophe could be. Improved exposure data is critical.”
To assess warehouse legal liability exposure, this means finding out not only the fluctuations in the values, but what type of goods are being stored, how they’re being stored, whether the warehouse is built to local standards for wind, earthquake and flood, and whether or not the warehouse owner has implemented any other risk mitigation measures, such as alarm or sprinkler systems.
“Since most models treat all warehouses equally, even if a location doesn’t model well initially, specific measures taken to protect stored goods from damage could yield a substantially different expected loss, which then translates into a very different premium,” Evans said.
That extra information gathering requires additional time but the effort is worth it in the long run.
“Better understanding of an exposure is key to strong underwriting — and strong underwriting is key to longevity and stability in the marketplace,” Evans said.
“If a risk is not properly understood and priced, a customer can find themselves non-renewed after a catastrophe results in major losses — or be paying two or three times their original premium,” he said. Brokers have the job of educating clients about the long-term viability of their relationship with their carrier, and the value of thorough underwriting assessment.
The Model to Follow
So the question becomes: How can insurers begin to elevate location level modeling in the marine space? By taking a cue from their property counterparts and better understanding the exposure using better data, science and engineering.
For stored goods coverage, the process starts with an overview of each site’s risk based on location, the construction of the warehouse, and the type of contents stored. After analyzing a location, underwriters ascertain its average values and maximum values, which can be used to create a preliminary model. That model’s output may indicate where additional location specific information could fill in the blanks and produce a more site-specific model.
“We look at factors like the existence of a catastrophe plan, and the damage-ability of both the warehouse and the contents stored inside it,” Evans said. “This is where the expertise of our engineering team comes into play. They can get a much clearer idea of how certain structures and products will stand up to different forces.”
From there, engineers may develop a proprietary model that fits those specific details. The results may determine the exposure to be lower than originally believed — or buyers could potentially end up with higher pricing if the new model shows their risk to be greater. On the other hand, it may also alert the insured that higher limits may be required to better suit their true exposure to catastrophe losses.
Then when the worst does happen, insureds can rest assured that their carrier not only has the capacity to cover the loss, but the ability to both manage the volatility caused by the event and be in a position to offer reasonable terms when renewal rolls around.
For more information about Berkshire Hathaway Specialty Insurance’s Marine services, visit https://bhspecialty.com/us-products/us-marine/.
Berkshire Hathaway Specialty Insurance (www.bhspecialty.com) provides commercial property, casualty, healthcare professional liability, executive and professional lines, surety, travel, programs, medical stop loss and homeowners insurance. The actual and final terms of coverage for all product lines may vary. It underwrites on the paper of Berkshire Hathaway’s National Indemnity group of insurance companies, which hold financial strength ratings of A++ from AM Best and AA+ from Standard & Poor’s. Based in Boston, Berkshire Hathaway Specialty Insurance has offices in Atlanta, Boston, Chicago, Houston, Los Angeles, New York, San Francisco, San Ramon, Stevens Point, Auckland, Brisbane, Hong Kong, Melbourne, Singapore, Sydney and Toronto. For more information, contact [email protected].
The information contained herein is for general informational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any product or service. Any description set forth herein does not include all policy terms, conditions and exclusions. Please refer to the actual policy for complete details of coverage and exclusions.
This article was produced by the R&I Brand Studio, a unit of the advertising department of Risk & Insurance, in collaboration with Berkshire Hathaway Specialty Insurance. The editorial staff of Risk & Insurance had no role in its preparation.