I was gratified to see that the editors of the New York Times had come around to the strategy I arrived at by inspection for managing whatever threats might be posed by artificial intelligence:
When companies create products that hurt people, the victims can sue. The legal system allows those who are harmed to seek compensation from those who are responsible. These liability laws are one of the oldest forms of corporate regulation, and an effective one. Liability is a powerful incentive to behave responsibly. It speaks to corporations in a language they understand: Money.
Artificial intelligence is an industry in need of incentives to behave responsibly, to judge by the litany of transgressions to which leading A.I. companies have confessed in recent months. Their programs have hacked or attempted to hack the databases of private companies and public institutions, including the governments of the United States and Australia.
This is a welcome addition:
Liability is an incomplete answer for the challenges posed by A.I., but it has two great advantages: The laws are on the books, and they are powerful. Makers of cars, airplanes, tobacco products and opioids have all been forced by lawsuits to change how they do business. They made their products safer, restricted access to them or both. Most recently, the social media company Meta agreed to pay billions of dollars to states and to start changing its policies toward children.
and
State legislatures have a role to play as well. They can write statutes to clarify the application of existing laws to A.I. Legislatures can also hold A.I. companies to a higher standard of accountability than current law does, given the risks. Companies are generally liable if a plaintiff can show that they knew or should have known about the danger their products posed. But some particularly dangerous activities, like storing toxic chemicals and operating a nuclear power plant, are held to a standard of strict liability, meaning that the company is responsible for harm even if it had no knowledge of the danger in advance.
The theory of strict liability is simple: If you own a tiger, and the animal bites a neighbor, the essential fact is that you own a dangerous wild animal. It doesn’t matter how cleverly the tiger was trained, or how carefully it was guarded. Bryan Choi, a law professor at the University of Colorado, notes that in the early days of aviation, lawmakers imposed this strict liability standard on the dangerous new technology. In the case of A.I., top executives themselves have spoken publicly about the dangers, up to the possibility of human extinction.
Some industry advocates argue that enforcing liability laws will stifle innovation, because A.I.’s defining breakthrough is its ability to act independently. Indeed, they argue that any significant restraints on A.I.’s development would be a mistake because the potential benefits of racing ahead are so great. In a recent interview, Sam Altman, the chief executive of OpenAI, said, “We believe that the world should accept some bad things happening for the benefits of this technology.”
Yet other industries do not escape legal responsibility for their products in the name of innovation.
Strict liability is essential for LLM AI. Due to the very nature of the technology it shouldn’t be necessary to prove intent or negligence; only harm and causation. Those hurdles will be high enough. If a company deploys a system whose particular harmful behavior cannot reliably be anticipated even with reasonable care requiring victims to identify a negligent act can leave them bearing the costs of a commercial activity from which others profit. That is a substantive argument for strict liability, and one developed in legal scholarship on AI.
