LW - Investigating an insurance-for-AI startup by L Rudolf L

The Nonlinear Library: LessWrong

Link to original article
Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Investigating an insurance-for-AI startup, published by L Rudolf L on September 21, 2024 on LessWrong.
We (Flo & Rudolf) spent a month fleshing out the idea of an insurance-for-AI company. We talked to 15 people in the insurance industry, and did 20 customer interviews. We decided not to continue, but we think it's still a very promising idea and that maybe someone else should do this. This post describes our findings.
The idea
Theory of change
To reduce AI risks, it would be good if we understood risks well, and if some organisation existed that could incentivise the use of safer AI practices. An insurance company that sells insurance policies for AI use cases has a financial incentive to understand concrete AI risks & harms well, because this feeds into its pricing. This company would also be incentivised to encourage companies to adopt safer AI practices, and could incentivise this by offering lower premiums in return.
Like many cyber-insurance companies, it could also provide more general advice & consulting on AI-related risk reduction.
Concrete path
TL;DR: Currently, professionals (e.g. lawyers) have professional indemnity (PI) insurance. Right now, most AI tools involve the human being in the loop. But eventually, the AI will do the work end-to-end, and then the AI will be the one whose mistakes need to be insured. Currently, this insurance does not exist. We would start with law, but then expand to all other forms of professional indemnity insurance (i.e.
insurance against harms caused by a professional's mistakes or malpractice in their work).
Frontier labs are not good customers for insurance, since their size means they mostly do not need external insurance, and have a big information advantage in understanding the risk.
Instead, we would target companies using LLMs (e.g. large companies that use specific potentially-risky AI workflows internally), or companies building LLM products for a specific industry.
We focused on the latter, since startups are easier to sell to. Specifically, we wanted a case where:
LLMs were being used in a high-stakes industry like medicine or law
there were startups building LLM products in this industry
there is some reason why the AI might cause legal liability, for example:
the LLM tools are sufficiently automating the work that the liability is plausibly on them rather than the humans
AI exceptions in existing insurance policies exist (or will soon exist)
The best example we found was legal LLM tools. Law involves important decisions and large amounts of money, and lawyers can be found liable in legal malpractice lawsuits. LLMs are close to being able to do much legal work end-to-end; in particular, if the work is not checked by a human before being shipped, it is uncertain if existing professional indemnity (PI) insurance applies. People who work in law and law tech are also, naturally, very liability-aware.
Therefore, our plan was:
Become a managing general agent (MGA), a type of insurance company that does not pay claims out of its own capital (but instead finds a reinsurer to agree to pay them, and earns a cut of the premiums).
Design PI policies for AI legal work, and sell these policies to legal AI startups (to help them sell to their law firm customers), or directly to law firms buying end-to-end legal AI tools.
As more and more legal work is done end-to-end by AI, more and more of the legal PI insurance market is AI insurance policies.
As AI advances and AI insurance issues become relevant in other industries, expand to those industries (e.g. medicine, finance, etc.).
Eventually, most of the world's professional indemnity insurance market (on the order of $10B-100B/year) has switched from insuring against human mistakes to insuring against AI mistakes.
Along the way, provide consulting services for countless business...