Machina Economica, Part II: The Commodification of Risk
Phillip Moreira Tomei
This is the second entry in a series on AI integration in the economy, exploring its financial, sociopolitical and historical implications. This entry focuses on economic risk and its financialisation.
Aivazovsky, Ivan. Storm Seascape Shipwreck. 1860
The pricing, parcelling and selling of risk has enabled the modern world. From the first maritime insurance policies in mediaeval Genoa to the complex catastrophe bonds of today, our ability to quantify, distribute, and mitigate uncertainty has fundamentally shaped global commerce and societal resilience. Like the globalisation of trade then, the artificialisation of intelligence is the nascent infrastructure of our economy. With it, we are faced with implementing probabilistic systems at the heart of a fragile economic superstructure. A robust risk market around AI can provide essential market correction. Our approach to the uncertainty of the $600bn question will play a crucial role in shaping the collective future of the machine age.
Resicum, the ancestor to the English risk, is a neologism; it first occurs in a proto-insurance document recorded by Geonese traders on the 26th April 1156. The contract allocates the resicum to the investor in a shipment, in doing so the danger and uncertainty of the voyage was borne by those safe on land. Maritime commerce was enormously profitable in the early modern period but also highly risky, piracy was widespread and the prediction of fatal storms impossible. Prior to the introduction of the resicum, sea voyages were a high-stakes gamble for captains and their crews, they alone bore the full weight of potential dangers while also being the sole beneficiaries of any profits. The resicum, however, revolutionised this dynamic by distributing both the possible gains and losses across a wider group of stakeholders. As Karla Mallete says ‘resicum put a number on contingency, and in so doing it rationalised risk’.
Modern insurance policies, like so many innovations in the business of risk, also began with a catastrophe. The Great Fire of London devoured 13,000 houses in 1666. Economist Nicholas Barbon established the first fire property insurance company by insuring 5,000 homes. Each of the newfound insurance companies employed their own fire department to minimise damages to the properties they covered. Together with the financialization of risk came the benefits of its aligned incentives, the institution of risk mitigation as enterprise.
Selling Catastrophe
The final step in this story was the transmutation of risk into a commodity, yet again with a disaster. On August 26th 1992, Hurricane Andrew struck the Gulf coast and western Florida, inflicting $27 billion in damages, of which $15.5 billion was covered by insurance. Andrew's impact led to the insolvency of eight insurance companies, threatening the stability of the industry, necessitating a comprehensive reevaluation of the financialisation of risk. What was needed was an institutional arrangement designed to enhance insurance capacity and distribute risk more effectively.
The catastrophe or CAT bond allowed financialised risk to be commodified and transferred to investors. Through tax-haven based Special Purpose Vehicles,1 CAT bonds enable diverse investors - from Japanese pension funds to Swiss private banks - to place significant wagers on events like Bangladeshi floods or Australian wildfires by owning a security linked to a catastrophe. Since natural disasters are generally considered independent of economic factors, these investments are uncorrelated with traditional markets, offering portfolio diversification opportunities. There is an element of recursion, as CAT bonds represent a stake in risk yet are themselves used to de-risk portfolios over-exposed to market fluctuations.
Polacek, Andy. "Catastrophe bonds: A primer and retrospective." Chicago Fed Letter 405 (2018)
CAT bonds operate as follows: an institutional investor purchases a bond with a specific triggering event such as ‘West Coast Earthquake’, if the specified event is triggered as specified (often certain thresholds must be met) the investor loses their principal, which is used to cover damages. However, if no triggering event takes place, the bondholder receives their initial investment plus a substantial return funded by the insurer or reinsurer
Speculators now have the opportunity to diversify their portfolios by engaging in "event-linked securities." These typically involve 1-3 year contracts covering a wide range of risks, including hurricanes, droughts, terrorist attacks, wars, cyber threats, civil unrest, industrial accidents, and water shortages. A notable feature of these securities is the use of "parametric" triggers. These mechanisms link payouts to specific, measurable real-world events. As a result, CAT bonds effectively create a monetary and time-based representation of the underlying demographic, geophysical, or epidemiological landscapes, encompassing all their complex potential outcomes.
As the world becomes enraptured with AI failure, of both the financial and catastrophic kinds, the CAT bond provides a way to cushion the uncertainty in implementing an inscrutable ‘black box’ technology into the heart of our socioeconomic systems. While we may not be able to insure the end of the world we can provide the funds to rebuild after collapse. In doing so we disincentive haphazard implementation and loss of human oversight over machinic economic value production. CAT bonds for AI Risk and the modelling that will come with them also provide a market-based solution to overzealous AI investment while providing prosocial benefits.
As Software Eats the World, Insurance Saves the World
As software has become necessary for the operation of every major business, so has cyber risk become the most salient form of business risk. The cumulative cost of cyberattacks increases by 15% every year. Insurance stands as one of the most promising solutions for addressing this insecurity. A well-developed market for cyber incident insurance could, among other benefits, provide financial incentives for organisations to improve their cyber hygiene, thereby reducing cyber risk for society overall.
Yet cyber insurance remains woefully under-realised, only 13% of businesses are covered adequately. Most cyber insurers offer only low coverage and high deductibles. Pricing and risk modelling are in its infancy and policies are hard to price. Technology is constantly changing and threat actors are continuously innovating.
Yet insurance is not just a financial tool for businesses, it is crucial infrastructure in an uncertain world. Just as seafaring merchants relied on insurance in the face of piracy, storms and mutiny. Insurers are key aggregators in the information ecology, their birds eye view of industry risk and proprietary data makes them the idealised knowledge actors in the Keynesian sense. Premium prices are ultimate informational compression.
Armed with this information, insurance reduces the risks it underwrites. In cyber, insureds have started collaborating with their insurers. New generation firms such as Coalition, incentivise their policyholders to take cybersecurity measures by reducing premiums and offering favourable rates according to the level of visibility Coalition is given to the insureds’ cyber infrastructure. The more visibility they have over their policyholders the more accurately they can price a policy and the better understanding they have of the global risk picture.
Coalition calls this ‘Active Insurance’, the idea being that instead of only being in touch when a claim is made they are deeply involved with the reduction, assessment and incident response of a client’s cyber infrastructure. While this may be heralded as an innovation in LinkedIn circles, it is no different than the insurance-run fire brigades and inspectors of 17th century Britain.
The Artificalisation of Intelligence, the Artificialisation of Risk
As Gary Zhexi Zhang notes: ‘Insurance is a kind of existential rent, the cost of holding a world in place against an uncertain but seemingly calculable future.’ Just as the risk of storms and the returns of trade was accounted for by the insurance societies of mediaeval Venice, so may the deployment of AI in the planetary infrastructure. Both the weather and the high-dimensional vector spaces of transformer-based models, are chaotic, dynamical systems we must grapple with in order to transact, construct and live.
As AI integration into economic activity becomes a competitive necessity it is not a stretch to imagine that just like every company has had to become a software company, so will every company become an AI company. The CrowdStrike bug that took down 8.5 million computers and thousands of aeroplanes, tv channels and shipping ports in July 2024 illustrates the systemic risk of software eating the world. The potential losses of AI failure when it is embedded in the software that runs global manufacturing, supply chain management and financial systems is of growing concern. The more interconnected the world is, the more consequential single points of failure can be, AI as a form of compression will only accelerate this process.
Central banks, the guardians of macroeconomic risk, are taking note, venture capital analysts project AI Risk management to become a $276bn business by 2030, a cottage industry of evaluators, consultants, auditors and assurance tech is already arising. Insurance and insurance linked securities (CAT Bonds) will ultimately abet, grow and price this world, financialising AI Risk. Engendering the anti-speculative framework that will guide cost-benefit analysis of AI deployment in the real world.
If cyber risk understanding is in its infancy, AI risk modelling is still gestating. We have yet to see the modes of failure that this novel piece of the planetary infrastructure will yield. Current research focuses on AI risk in the model itself, divorced from its deployment as a domino, in the physical and economic fabric of the world. As the technology matures, so will the research community need to move from looking at AI failure in silico to in vivo.
Only then can the financial and socialised cost of a probabilistic critical infrastructure be understood. We yet again face a clash between our determination and the stochasticity of our world.
Notes
- A special purpose vehicle is a legal entity that allows multiple investors to pool their capital and make an investment in a single company while isolating financial risk. ↩︎
