The global insurance industry has yet to get to grips with the risks it could face from the growing use of artificial intelligence, according to experts.
The use of AI, from factory robots to automated processes to digital assistants, is growing across a variety of industries. Such technology is seen as the key to unlocking substantial productivity gains and cost savings, but if systems go wrong, they can disrupt services or give bad advice, requiring compensation payments to customers, or even cause property damage or bodily injury.
The claims could be large: In August 2012, stockbroker Knight Capital Group lost $440 million after a series of erroneous trades executed by its computer systems, while other examples include a 2017 systems outage at British Airways, which forced parent company International Airlines Group to book a €65 million provision in its full-year 2017 accounts. Similar failures at fellow airline Delta in August 2016 and December 2017 cost the company roughly $150 million and $40 million, respectively.
While not strictly AI failures, those incidents demonstrate the losses that can occur through companies’ dependency on technology systems. Neil Beresford, a partner at law firm Clyde & Co., argued that there are some parallels between AI and so-called silent cyberrisk, a term that refers to standard policies being exposed to cyber claims without their being explicitly included or priced for.
At a seminar covering the impact of AI on underwriting in July, Beresford said London underwriters would need to scour policies for "hidden" AI exposures in the same way they had for silent cyber risks, because AI could affect a broad range of coverage types in ways that underwriters had not foreseen.
"There is a need for an understanding of what these products do, how they work, how they fit into the wider picture and the sorts of losses they could cause," Beresford told S&P Global Market Intelligence in an interview.
He added that he had not yet seen a tightening of policy wording around AI: "I don't know if that is because underwriters are looking at these things and understanding them or whether, as I suspect, [they are] not always appreciating the unusual nature of the risks to which these products give rise."
Although Beresford said the size of the problem to insurers is difficult to quantify, he pointed to China's ambitions, which include the creation of an AI industry with an output value in excess of 1 trillion yuan by 2030, to indicate the potential scale and rate of development.
Neil Hare-Brown, founder and CEO of cyber risk advisory firm Storm Guidance, said AI was a "talking point" but that "from a risk perspective, it's just not even on insurers' radars yet."
Max Richter, European insurance analytics practice lead at consulting firm Accenture, added that he had seen a "strong response" from the industry in changing wordings to tackle cyber risks, but that despite some evidence of "early discussions" on AI, "the action on that is still relatively limited." He added that the insurance industry's awareness of AI risks was mixed, with some providers starting to think about what they needed to evaluate risks properly and others "less informed."
Rise of the robots
Risks from automated machinery, such as factory robots and driverless cars, are less likely to catch the industry unawares because they are more likely to have their own tailored insurance policies or wordings. Richter said: "There will be a very direct discussion between [insurers] and the manufacturers of autonomous cars about the nature of the risk and who carries the liability, and everyone is going into it with their eyes fully open."
The real potential for unforeseen risks lies where technology is being used to automate business processes or entire businesses. Insurers could be caught out because they are still getting to grips with how AI works.
Richter said: "[Insurers'] ability to judge whether a third-party company has gone through the right processes and applied the right controls in deploying their AI is still at a relatively limited level, but that is a capability that needs to be increased as we go into the future."
To err is human...
Although he is unconvinced that insurers face hidden exposures from AI, Graeme Newman, chief innovation officer at technology-focused underwriting agency CFC Underwriting, acknowledged that where a human conducting a process may make one or two mistakes, a computer can make mistakes "at a huge rate."
"In the old days when things were much more manual, things would go wrong on a smaller scale," he said, "and equally if there are technology providers that sit across a number of different [firms] in an industry, clearly that presents a bigger aggregation problem for an insurer because one failure can impact multiple, and often thousands, of accounts."
But Newman said insurers' awareness of AI risks is growing, in part because they themselves are using it in processes such as claims triage, and in part because of the rise of insurtech companies helping analyze data and risk in new ways. "I don't think it is something the industry is unaware of or sleepwalking into," he added.
And insurance brokers are trying to minimize confusion about what is and is not covered. Bill Boeck, insurance and claims counsel at insurance broker Lockton Cos., said: "If there is some room for doubt about whether the use or implementation of artificial intelligence is covered, then we will generally try to craft policy language that will specifically address it."
As of Aug. 16, US$1 was equivalent to 6.88 Chinese yuan.