INSURING AI: THE ROLE OF INSURANCE IN ARTIFICIAL INTELLIGENCE REGULATION.

• Author: Lior, Anat

TABLE OF CONTENTS I. INTRODUCTION 469 II. ACTUARIAL SCIENCE MEETS AI 474 III. THE PROBLEM WITH AI'S INSURABILITY 479 IV. INSURANCE AND EMERGING TECHNOLOGY 481 A. Insurance of New Technologies 482 B. Insurance and Its Assimilation of Innovations 484 C. Current Suggestions for Insuring AI 487 1. "Turing Registry" 487 2. In-House Insurance 489 3. The Automated and Electric Vehicles Act and Road Traffic Act 490 4. Manufacturer Enterprise Responsibility ("MER") for Autonomous Vehicles 491 5. National Insurance Fund for Autonomous Vehicles 493 6. European Parliament Compensation Fund 496 V. INSURANCE AS A CIVIL REMEDY FOR AI DAMAGES 498 A. The Disadvantages 499 1. Moral Hazards and Insurance 499 2. The Lack of a Global Adoption of Insurance in the AI Context 501 3. Cost Allocation and Premium Estimation 502 a. Known Unknowns 502 b. The Feasibility of Insurance in the AI Market 504 4. Exceptions and Exclusions 509 B. The Advantages 511 1. Regulation by Liability Insurance: Behavior Channeling 511 2. Questions of Liability and Predictability 519 C. Potential Liability Barriers and the Role of Insurance 520 1. Joint Causation 520 2. Hacking AI Entities 522 VI. A PROPOSAL FOR INSURANCE IN THE AI REALM 524 A. Building Upon Existing Infrastructure 524 B. What About the Singularity? 527 VII. CONCLUSION 529 I. INTRODUCTION

At its core, "[i]nsurance is a business that specializes in risk management." (1) Although insurance is mainly viewed as an ex post tool for indemnification of disasters through risk pooling, loss spreading, risk reduction, and shifting, (2) insurance serves other important roles of risk reduction and management ex ante. In today's technological age, nothing requires more risk reduction and management than the emerging technology of AI commercial machines, robots, agents, and algorithms (hereinafter "AI entities" (3)), which are increasingly integrated into our everyday routine. (4) Usually, insurance companies can offer policies to cover different types of damages since they "know a thing or two" because they have "seen a thing or two." (5) This is not necessarily the case in the emerging field of AI.

So far, the discussion about AI and insurance has followed two distinct inquiries. (6) The first, and more prominent, concerns how AI influences actuarial science and the operation of the insurance market. This conversation has largely centered on using technology, in this case AI, in the insurance sector "to create savings and efficiency in the insurance value chain." (7) This optimistic thread of AI literature expects the massive calculating power of AI to yield increasingly accurate premiums, revolutionizing the insurance industry. (8)

The second inquiry, which has yet to receive sufficient attention, considers not how AI will change the insurance industry but how the insurance industry can change AI. This inquiry considers insurers' ability to offer coverage when AI entities cause harm to humans or property. (9) This Article delves into this second problem and the important legal questions arising from the topic's intersection with emerging technologies, tort law, and insurance.

AI liability has been a prominent topic during the last several years, with scholars suggesting various liability regimes that should apply to AI-inflicted damages. (10) However, these scholars have failed to recognize the role of insurance in regulating AI-inflicted damages. (11) It is impossible to fully unpack AI liability without also discussing the insurance market, which has the potential to enable the AI industry to flourish despite its inherent unpredictability.

This Article will discuss the best way to utilize the existing insurance infrastructure to regulate current and future uses of AI entities. It argues that existing insurance infrastructure can be instrumentalized to regulate AI entities without need for a special AI insurance policy. Extending existing mandatory insurance compensation schemes, or "coercive insurance" (12) (i.e., no-fault accident compensation schemes) to AI activities will help bypass legal problems of liability and blameplacing, make more predictable which entity will compensate victims for damages, (13) and minimize insureds' dangerous behavior by mitigating moral hazard risks and requiring certain safety standards be met to maintain the validity of an AI insurance policy. (14) The existing insurance practices will adapt to the challenges presented by businesses using AI. Insurance companies underwrite policies for businesses all the time. Even though these businesses already use AI, this usage will not require creating new policies, only the adjustment of existing ones. And in situations where insurance is not legally mandated, users of emerging AI technology will still seek to buy optional insurance policies in order to hedge risks. (15)

The framework offered in this Article stems from the regulatory effect of liability insurance. (16) Although this Article does not support enforcing a comprehensive mandatory insurance compensation scheme for businesses manufacturing and using AI, it is unavoidable that the vast majority of businesses with a significant effect on the public will purchase liability insurance in the long run. This is because it is highly likely that they will be sued, so they will need to acquire some sort of hedging mechanism, either because courts will impose strict liability or out of fear that they will do so. (17) While this Article does not take a position on the appropriate liability regime for AI, it is important to note that a system of liability insurance requires liability rules, (18) otherwise businesses will not purchase insurance policies. The best way to make sure that insurers have the ability to regulate these businesses is to impose strict liability on accidents involving AI; (19) this liability regime will encourage businesses using AI to purchase policies. To receive insurance compensation, businesses must alter their behavior to qualify under a no-fault accident compensation scheme. This creates two important benefits. First, businesses will adopt improvements required by their insurers following a "reasonable person" standard. Second, businesses will adjust their activity level to ensure they are operating in a safe and efficient manner that will enable them to pay for damages that they may incur under their insurance policy.

But not all AI accidents can be prevented by adjusting activity levels. When these inevitable accidents do occur, they have tended to receive significant attention across media outlets, giving rise to a debate about who should be held accountable both within the legal realm and outside of it. (20) The tort system possesses immense power to publicize high-profile cases involving new dangers that may potentially lead to their elimination or reduction. (21) Such is the case with autonomous vehicle ("AV") accidents, which have caused anxiety to drivers and pedestrians around the world. However, when it comes to new and unexpected technologies, which gain publicity by cherry-picking the most egregious cases, this negative attention may lead to the loss of an important technology that possesses enough advantages to outweigh its flaws. Insurance can provide a way to avoid this fate.

The concept of using insurance to manage damages caused by AI entities has been discussed in the UK and EU. In 2017, the European Parliament published a resolution recommending a set of civil law rules for robotics. (22) Article 59 of that resolution discusses different insurance principles that can apply to AI entities. These include establishing a compulsory insurance scheme for specific categories of robots (including AI entities) and establishing a compensation fund. (23) The usage of insurance to manage AI entities is already a reality in the UK's automobile industry, where the Automated and Electric Vehicles Act was enacted in 2018. (24) This Act applied the existing insurance infrastructure of non-autonomous vehicles to autonomous ones. (25) These two examples demonstrate potential ways insurance can be used to regulate AI. Legislators must consider whether to extend existing insurance infrastructure to AI, as was implemented in the UK, or develop new regulatory schemes, as was suggested in the EU. This Article expands on the debate and ultimately endorses the UK route of using existing insurance infrastructure to regulate AI.

Creating a novel insurance scheme specifically for AI entities would be impractical because regulators and insurers currently lack the knowledge, resources, and time to form new insurance policies before AI entities become an even more integral part of our commercial lives. (26) Thus, the solution for AI-inflicted damages will have to stem from our existing insurance practices and infrastructure, which, given this technology's substitution effect, already cover most of the activities that AI entities will eventually perform. (27)

The existing insurance infrastructure offers stability and cohesiveness in regulating AI entities. Although a no-fault accident compensation scheme for AI entities, or any other voluntary AI insurance policy, has some drawbacks, these drawbacks are not unique to AI--they have arisen in the past whenever new technologies have emerged. History demonstrates the insurance industry's capacity to assimilate such emerging technologies. Moreover, the insurance industry is better placed to tackle this challenge than it has ever been: insurance companies will be able to use the forces of AI itself to better set premiums and identify the most efficient entity to make responsible for purchasing a liability policy. Here the two scholarly discussions about AI and insurance converge. By enhancing the art of risk assessment, AI helps to better regulate its own risks. (28)

This Article continues as follows. Part II introduces the unique features of AI entities in the insurance context. Part III focuses on...