Hermer demands urgent law to control artificial intelligence

The story frames artificial intelligence as an unregulated frontier, and Richard Hermer’s complaint that it is governed less strictly than a refrigerator is meant to shame us into building fences. But look at what is actually being fenced already. The knowledge of how these systems behave, their weights and training data and the results of the tests run on them, sits inside a handful of companies. That knowledge was assembled from a vast commons of public writing, shared code and openly published research. The question his framing skips is who will hold the keys once the law arrives, and whose access the new fence removes.

I do not quarrel with his alarm. He is right that the public has been left out of a decision that may shape every life after it, and right that “losing control of the future” is not melodrama. But his refrigerator deserves a closer look than he gives it, because it holds the answer his argument needs.

Why is a refrigerator safe? Not because one firm was licensed to make them and the rest forbidden. It is safe because what makes it safe is common property. The electrical standards are published. The energy label is printed on the door so that a woman in a shop can compare two machines with her thumb on the sticker. Any engineer can open the back panel and know what she is looking at, and any repairer in any town can find a fault and report it. The fridge is governed less by a ministry than by an enormous web of cooperation among the testers, inspectors, tradesmen and buyers who share one body of knowledge about what a safe fridge is. When the refrigerator did threaten the whole of humanity, through the chemicals in its cooling coils thinning the ozone layer, the international law that followed worked because the measurements of the damage belonged to everyone. Scientists in many countries pooled observations no government could suppress, and the treaty rested on that shared evidence. Hermer’s “robust international law” has a precedent, and the precedent is a commons.

Now consider the strongest case against me, which is serious. Some capabilities are dangerous, the argument runs, and spreading them openly hands them to the reckless. Better that a few accountable firms, licensed and supervised by states, hold the most powerful models behind locked doors where inspectors can watch them. Concentration, on this view, is not theft but containment.

This argument has been made by every party that expected to end up holding the key. It assumes that danger lives in the many hands and safety in the few. The record of this technology suggests otherwise. The pressure that worries Hermer comes from a small number of firms racing each other for a prize too valuable to share, cutting corners because the rival might get there first. A licensing regime built around those same firms does not remove the race. It removes the other runners and the spectators. The labs gain a moat drawn in statute, the ministries gain a short list of companies to telephone, and the public loses its witnesses: the independent researchers, the university groups, and the small builders who found flaws precisely because they could look inside. The future would not be brought back under control. Control of it would be certified as belonging to the few who already have it.

There is a comedy here that deserves to be said plainly. We are told the machine is too dangerous for the public to inspect, so it will be inspected by the people selling it, under rules partly drafted by the people selling it, and the public will be reassured by a press release. No refrigerator manufacturer has ever been granted that arrangement. If one asked for it, we would assume the fridge was leaking.

So Hermer’s urgent public debate should begin with a question his comparison raises but does not ask, which is what the law will make common and what it will make private. Regulation that requires published evaluations, open access for independent testers, shared reporting of failures, and standards anyone can read would govern AI the way we govern the fridge, through a commons of knowledge that many hands maintain. Regulation that licenses a club and seals its findings would be enclosure that calls itself safety.

Cooperation of this kind is hard. Shared testing is slow and underfunded, and it is often ignored until a disaster makes it fashionable. But it is the only form of oversight that has ever kept pace with an industry, because it scales with the number of people who care.

Somewhere tonight a repairer kneels on a kitchen floor, unscrews a back panel, and reads a wiring diagram printed on the inside for anyone who opens it. Nobody had to ask permission to print it there.