The artificial intelligence technology race, presents as much catastrophic risk for insurers as hurricanes, wildfires, floods, and earthquakes. Despite the fact, artificial intelligence is in its infancy, large corporations, investors and entrepreneurs are rushing these products, such as Chat GPT-4 (GPT-4), to market. It is impossible not to be in awe of these artificial intelligence programs for what they can do now and their promise for the future. While all Infants make mistakes, infants that appears to have the power of mythological gods that engender blind trust are particularly dangerous. If the past provides any guidance, the insurance and reinsurance industry will find itself compensating the injured and remediating damage as artificial intelligence grows up. The teen age years will be particularity challenging.
Today’s insurers, like those in 19th and mid-twentieth centuries, stand on the threshold of profound changes in the nature of risks to be addressed and the legal architecture built to determine and allocate liability for these new risks. In the midst of the Industrial Revolution, the insurers of the 19th and mid-20th Century sold occurrence and accident policies predicated on simple negligence theories of liability, a narrow interpretation of the definition of an occurrence and accident, and a reliance on privity. Likewise, manufacturers, transportation companies, and other businesses in that era (understandably) failed even to approximate the costs of doing business such as the number of workers and third parties killed or the extent of property damage. They also paid little regard to the various externalities such as air and water pollution from harmful waste disposal practices. Despite all these damages that were at least arguably occasioned by negligent practices it was hard for injured parties to perfect their claims in tort. The 19th and early 20th Century injury law heavily favored manufacturers and businesses. As a result, insured losses were small, and reinsurers treated the casualty business as a loss leader.
However, a legal revolution was brewing in the wake of the Industrial Revolution and its legacy of uncompensated injury and damage. Legal scholars and policyholder lawyers began pushing for change. This century-long struggle culminated in the American Law Institute’s 1965 publication of the Restatement of Torts, Second (402A), which discussed the imposition strict liability for defective products.
The Restatement of Torts, Second (402A) represented a watershed moment in the history of the insurance and reinsurance industry. This Restatement and the subsequent legislative tsunami overwhelmed insurers and reinsurers with new theories of liability which resulted in billions of dollars of losses. Perhaps the most prominent consequence was the asbestos crisis of the 1980s which saw many insurers and reinsurers go bankrupt and disappear. Insurers and reinsurers were forced to adapt to a new world of strict liability resulting in a restructuring of their casualty insurance policies and their reinsurance treaties. The impact of the Restatement of Torts Second and the subsequent Restatement of Torts, Third: Products Liability published in 1998 continue to significantly influence contemporary insurance and reinsurance product development.
Like their predecessors, insurers and reinsurers today find themselves in the middle of a potential sea change as we enter the age of artificial intelligence. Insurers and reinsurers which issue accident or occurrence policies or losses- discovered reinsurance treaties will soon learn (if they have not already) that they have long-tail artificial intelligence-related liabilities embedded in their portfolio. Some insurers use what will likely turn out to be an inadequate analog model for the new technologies’ underwriting, product development, and pricing. Other insurers and reinsurers are already using advanced analytics, big data, and artificial intelligence to re-engineer their underwriting. Everyone will struggle to find artificial intelligence experts, technology experts, data scientists, and far-sighted coverage lawyers. Thus, this technological revolution will tend to exacerbate difference between well-resourced insurers and reinsurers and those not so well-resourced.
One big question that the courts (or legislatures) will need to address, and underwriters and product designers will need to anticipate, is whether AI-supported, designed, or powered solutions are to be considered products or services. This distinction is vitally important as products are subject to negligence and strict liability claims, whereas services (to this point) are subject only to negligence claims.
Companies frequently underestimate the complexity and difficulty of developing and deploying artificial intelligence programs. This March, Fortune reported that the failure rate on artificial intelligence projects is between 83% and 92%. The complexity and high failure rate of artificial intelligence projects is a complicated underwriting problem for insurers. This high failure project rate and its quality control implications suggest an even more challenging catastrophe risk. Underwriters must develop tools to
assess data quality, reliability, accountability, and fairness. Another large problem may be the lack of algorithmic transparency and an inability to explain the results. Even more problematic, large language models, like GPT-4, lie or hallucinate.
Open AI’s GPT-4 is an excellent example of the current generation of generative artificial intelligence programs. I have been using GPT-4 daily and it is amazingly helpful. GPT-4 is also addictive.
GPT- 4 is reported to have 170 trillion parameters. These trillions of parameters enable GPT-4 to handle multimodal data, both generating longer and more internally consistent text and images. Exhibiting what seems to be almost human like intelligence, it collaborates with you, provides research, makes unexpected connections, and creates polished drafts. Having passed a simulated bar exam with honors, GPT-4 can make users feel somewhat inadequate. On the other hand, the law firm Levidow, Levidow & Oberman, P.C. discovered Chat GPT can be a confident and authoritative liar. That firm was fined $5,000 after the court found one of its lawyers used Chat GPT to write a court brief that included false citations to non-existing case.
This fantastic technology comes with a universe of networked risks we don’t understand. Technology seduces you with brilliance, but like the serpent in the Garden of Eden, it can deceive you and lead you in entirely uncontemplated and undesired directions.
As these technologies mature, insurers will want to manage these risks and anticipate changes in injury law that may impact current underwriting techniques, policy wording, analytics, and pricing models. A well-designed and properly- priced reinsurance program, with the appropriate coverage grants, is a hedge against emerging technological and legal risk. While we do not practice law or give legal advice at Cobblestone Consulting LLC, I have spent forty years collaborating with coverage lawyers, insurers, reinsurance brokers and reinsurers developing enterprise risk management strategies and designing and negotiating broad reinsurance coverage grants that provide insurers with tools for managing risk. We can help.