Definition
In the context of insurance, risk is often described as an uncertainty of loss. AI risk refers to the uncertainty of loss that arises due to development, deployment or use of AI. A loss occurs due to a peril (cause or event) that results in damage (financial/reputational). AI risk occurs due to perils such as model underperformance, model unfairness and model copyright issue that results in direct financial loss to the AI company or their customers. A table of common AI perils and associated examples of losses that have occurred in recent years have been listed below.
We have compiled the list of all AI risks that can be insured here.
How is AI risk different from common insurable risks?
The actuaries and underwriters have depended on historical data of perils and their financial impact to calculate the premium of an insurance policy. This process is commonly referred to as class rating or manual rating. There is not enough historical data for perils due to AI, hence, insurers have to rely on different rating mechanisms referred as judgment rating.
Crucially, in the context of AI, it's worth noting that losses are likely to exhibit interdependence, a vital condition that makes insurability difficult. AI-related risks have the potential to swiftly disseminate globally, raising questions about the feasibility of geographically diversifying these risks. AI transcends geographical confines, and a more comprehensive understanding of the systemic repercussions of specific losses on the global economy is essential for enhancing the insurability of the aforementioned AI-related risks.
Are all AI risks insurable?
There are many AI risks that are not insurable such as AI going rogue or winner-takes-all (refer to the table by Brookings). The AI risks categorized as societal risks or ethical risks are also not insurable. Brookings also categorizes AI at three different levels of intelligence: narrow, general and super. Some of the risks mentioned in the table below such as destruction of society by Robots are only possible by general AI and super AI.
AI risk mitigation and management
There are five ways to manage risks - avoidance, reduction, retention, sharing and transfer. Performing AI Audits can lead to reduction in AI risks. Following good practices (scientifically or empirically proven) in different phases of AI development such as data collection, annotation, training, production setup and carefully building the underlying process leads to reduction in AI risk. Even after The AI risk can never be reduced to zero, hence, a risk transfer mechanism such as the financial guarantee provided by Atri Insurance Solutions and insured by big insurers becomes essential.
.png)
