It’s quite gratifying to see the argument I’ve been making for years gradually start seeping into common recognition. In this case it’s that LLM AI service providers (and their stockholders) should bear strict liability. Gabriel Rauterberg and Sarath Sanga remark in an op-ed at the Wall Street Journal:
The heads of frontier labs have said their companies could be responsible for such a mass tort, or worse. They’ve made the case against limited liability.
That case, as we argue in a recent paper, is strongest when a corporation’s profits and the risks it poses rise together, and when those risks are extreme and unbounded. That’s exactly the situation AI executives describe. Customers demand the capabilities that make the product so dangerous. The more successful the product, the more risk likely lands on the public.
Building more-capable models without these risks is the engineering problem that labs say they need more time to solve. Congress isn’t about to solve it. But Congress doesn’t have to. It can establish a liability rule and let incentives do the work.
In the 19th century, locomotives threw off hot embers that set nearby fields and barns ablaze. Lawmakers didn’t engineer a technical solution. They an
That should be easy for Congress to achieve even in a divided Congress.
Building better models is only one of the engineering problems AI service providers face. They must also find ways to deploy them safely. Strict liability for harms their services cause would give them a powerful incentive to do so. And if the potential harm exceeds what the company can pay, limiting the loss to the company’s assets leaves the public bearing the risk while shareholders retain the upside. That is the part of the authors’ argument Congress should take seriously and the sooner the better.
Meh. At the upper end of the wealth and power spectrum the legal system is pretty ineffective. At the executive level their income and wealth is insulated and there is close to zero chance of jail so the incentives lean heavily towards risk.
Steve
I think the historical analogies are a bit misleading, but they don’t seem to be arguing that existing court-made tort remedies are inadequate (such as negligence and strict liability), but the extent of liability could bankrupt the companies, limiting plaintiff recoveries.
That’s a real concern, but the usual remedy is to require insurance to guard against the risk of not being able to pay liabilities. (They mention asbestos, and while many of the key asbestos companies filed bankruptcy, dozens of lawsuits are still filed every week in (the metro-east) Illinois for claims on the asbestos funds, which are a majority financed from insurance payouts.) Still, insurance companies may seek to limit their exposure, and at the AI Apocalypse level, the courts won’t be functioning anyway.
The authors pivot back and forth between talking about investors and executives. Limited liability does not protect executives from personal liability for their own tortious conduct. Simple shareholders typically can’t be liable for any tortious conduct because they don’t actively participate in the management of the corporation. Their approach is to expand liability for those not involved in day-to-day operations, not expand the liability of those actually working in the company.
Since everyone reading this is probably invested in AI in one form or another, the cardinal sin here is a lack of follow through. Don’t worry dear reader, your 401k won’t make you personally liable, Congress can just pass a law that singles out the weirdo tech-bros that everyone dislikes. Two of the historic examples given imposed post-bankruptcy liability on shareholders proportionate to their ownership. Why is that not the rule that should be advocated here? I can think of two answers, broad shareholder liability would be politically unfeasible, and the other is that the argument was never principled to begin with.
One of the things about that, PD, which I agree would require changes in the law, would be that insurance companies would be highly incentivized to impose requirements on the AI service providers to limit their risk and the insurances companies are much better suited to assess, price, and limit risk than the Congress is.